Curative Analysis of Patients with Hepatocellular Carcinoma Using Transcatheter Arterial Chemoembolization Combined with Radiofrequency Ablation.

BACKGROUND Transcatheter arterial chemoembolization (TACE) combined with radiofrequency ablation (RFA) can improve the survival of patients with hepatocellular carcinoma (HCC). The purpose was to explore the characteristics of high-risk and low-risk groups of HCC patients receiving combination therapy using a decision tree model. MATERIAL AND METHODS This retrospective cohort study investigated HCC patients treated with a combination of TACE and RFA at our hospital from 2012 to 2018. Decision tree analysis was used to study the 1-year prognosis of patients, and patients were divided into high-risk and low-risk groups. RESULTS We included a total of 142 patients with HCC, 21.83% female and 78.17% male, with the median age of 60 years old. The median follow-up was 13.5 months; 39.44% of patients had progressive disease or death (high-risk group) and 60.56% of patients did not have progressive disease or survival (low-risk group). The area under the curve (AUC) of the decision tree model was 0.846. There were significant differences in sex (P=0.003), age (P=0.038), tumor number (P=0.043), number of RFAs in the first treatment cycle (P.

Background

Hepatocellular carcinoma (HCC) accounts for approximately 75–85% of primary liver cancers; it was the sixth most common cancer and the third leading cause of cancer-related death in 2020 [1]. In Western populations, 90% of patients progress from the chronic inflammatory fibrotic stage to cirrhosis, through low-grade and high-grade dysplastic nodules, and eventually to tumors [2]. In China, HCC has high morbidity and mortality due to the high incidence of hepatitis B [3,4]. Patients with HCC generally have no typical clinical symptoms in the early stage. Many patients are in the middle and late stages when they are diagnosed [5]. Moreover, HCC patients often also have various degrees of liver cirrhosis, liver insufficiency, and other manifestations of chronic liver damage, or lose the opportunity for radical surgical resection due to the location, number, size, and other related characteristics of the tumor, so the prognosis is very poor and the mortality rate is higher [6,7].

For the treatment of advanced HCC, transcatheter arterial chemoembolization (TACE) is currently widely used as an effective local palliative treatment [8]. Several studies have proved the curative effect of TACE on HCC and it is widely used in clinical practice, which could significantly improve the survival rate of patients with advanced HCC [9,10]. Radiofrequency ablation (RFA) technology is widely used in the treatment of HCC patients at various stages of disease due to its advantages of being minimally invasive, safe, convenient, and repeatable [11]. RFA provides similar odds of survival as liver resection or liver transplantation [12,13], but these latter 2 treatments are associated with local recurrence or a high recurrence rate [14,15]. At present, studies suggest the combination of TACE and RFA treatment for HCC patients to improve their survival and quality of life. However, many researchers have reported that in patients with HCC, TACE therapy combined with RFA therapy is associated with significantly better long-term survival compared with TACE or RFA therapy alone [16]. Few studies have explored the characteristics of the better-prognosis and poor-prognosis populations of patients treated with the combination of TACE and RFA [17-19]. It is important to identify high-risk groups receiving combination therapy and to intervene accordingly.

This study aimed to use machine learning methods to stratify HCC patients into high-risk and low-risk groups, and to explore the efficacy of TACE combined with RFA therapy in various populations.

Material and Methods

Study Design and Population

This retrospective cohort study enrolled HCC patients receiving TACE combined with RFA between 2012 and 2018 in the First Affiliated Hospital of Xinjiang Medical University. TACE combined with RFA treatment in our study was defined as performing TACE treatment for the lesions selected for treatment, followed by re-examination after 1 month, and then RFA adjuvant therapy was used. A total of 142 patients were included in this study.

Follow-up was performed approximately 1 month, 3 months, and 6 months after HCC treatment. If patients feel unwell during the follow-up period, they were encouraged to quickly seek medical attention. All patients were required to be followed up in hospital and undergo physical examination, including routine blood tests, coagulation function, liver and kidney function, tumor markers, and abdominal contrast-enhanced CT/MRI.

This study was approved by the Ethics Committee of the First Affiliated Hospital of Xinjiang Medical University.

Outcome Variable

The endpoint of this study was progressive disease or overall mortality of the patients within 1 year, who were defined as high-risk group. Progressive disease was defined as a 20% increase in the sum of the diameters of the viable target lesions or the appearance of new lesions.

Data Collection

Data on the demographic and clinical characteristics of the study population were collected. Demographic characteristics included age and sex. Clinical characteristics included tumor location, tumor number, tumor size, portal vein tumor thrombus (PVTT), hepatitis, number of TACEs in the first treatment cycle, number of RFAs in the first treatment cycle, alpha-fetoprotein (AFP) level (ng/ml), total bilirubin (Tbil) level (μmol/L), albumin level (g/L), alanine transaminase (ALT) level (U/L), aspartate transaminase (AST) level (U/L), lactic dehydrogenase (LDH) level (U/L), alkaline phosphatase (ALP) level (U/L), Child-Pugh score, Barcelona Clinic Liver Cancer (BCLC) stage, and China liver cancer staging (CNLC). One RFA treatment, regardless of the number of HCC lesions during the treatment period, was counted as 1 RFA.

Statistical Analysis

Quantitative data were tested for normality using the Kolmogorov-Smirnov test. Normally distributed data were described as mean±standard deviation (mean±SD), and the t test was used for comparison between 2 groups. Nonnormally distributed data were described by the median and interquartile range (M [Q1, Q3]), and comparisons between groups were performed using the Mann-Whitney U rank-sum test. The categorical data were described by the number of cases and the constituent ratio (n [%]), and the chi-square test was used for comparison between groups. In the variables of the study population, there were 10 patients with missing AFP, 2 patients with missing Tbil, 1 patient with missing ALT and AST, and 4 patients with missing NAD and ALP. Missing values were imputed using multiple imputations. We used the data before and after imputation to compare the differences between the 2 groups to explore if there was a statistically significant difference between the 2 groups. We used LASSO regression to screen out variables that were statistically significant for the outcome and then the screened variables were entered into the decision tree for 5-fold cross-validation. The results predicted by the decision tree were divided into low-risk groups and high-risk groups according to the cut-off point. We compared the characteristics of the low-risk and high-risk groups. The cumulative incidence of progressive disease or death within 1 year was plotted for the high- and low-risk groups, and association analyses were performed to explore whether the groups we differentiated were independently associated with progressive disease or death within 1 year. Model 1 did not adjust for any confounders, which was a rough model. Model 2 adjusted for the number of TACEs during the first treatment and LDH level. Model 3 adjusted for variables not used to build the model, including the number of TACEs during the first treatment, LDH, tumor location, tumor size, PVTT, hepatitis, Child-Pugh score, BCLC, and CNLC.

We used SAS v. 9.4 (SAS Institute, Cary, North Carolina) and Python software v. 3.7.4 (Python Software Foundation, DE, USA) for analyses.

Results

Characteristics of the Study Population

The characteristics of the studied patients are shown in Table 1. We found that 60.56% (86/142) of patients who received the combination of TACE and RFA were free of progressive disease or survived within 1 year, and 39.44% (56/142) of patients had progressive disease or death within 1 year. The median age was 60 years and 21.83% (31/142) patients were female and 78.17% (111/142) were male. Most of the patients (79.58% [113/142]) had hepatitis. The median AFP was14.35 ng/ml, the median ALT was 25.36 U/L, and the median AST was 30.60 U/L. Most patients (48.59% [69/142]) were BCLC stage A, and 38.03% (54/142) were CNLC stage Ia.

Table 1

Comparison of low-risk and high-risk groups receiving combined TACE and RFA therapy.

Variables	Total (n=142)	Low-risk (n=86)	High-risk (n=56)	Statistics	P
Sex, n (%)				χ2=9.020	0.003
Female	31 (21.83)	26 (30.23)	5 (8.93)
Male	111 (78.17)	60 (69.77)	51 (91.07)
Age, Mean±SD	59.90±10.57	61.38±10.19	57.63±10.83	t=2.09	0.038
Tumor location, n (%)				χ2=0.985	0.611
Right liver	93 (65.49)	59 (68.60)	34 (60.71)
Left liver	17 (11.97)	9 (10.47)	8 (14.29)
Bilateral liver	32 (22.54)	18 (20.93)	14 (25.00)
Tumor number, n (%)				χ2=6.271	0.043
One	98 (69.01)	61 (70.93)	37 (66.07)
Two	21 (14.79)	8 (9.30)	13 (23.21)
Three or more	23 (16.20)	17 (19.77)	6 (10.71)
Tumor size, n (%)				χ2=1.487	0.685
≤3 cm	9 (6.34)	7 (8.14)	2 (3.57)
3–5 cm	54 (38.03)	33 (38.37)	21 (37.50)
5–10 cm	76 (53.52)	44 (51.16)	32 (57.14)
>10 cm	3 (2.11)	2 (2.33)	1 (1.79)
PVTT, n (%)				χ2=0.024	0.876
No	125 (88.03)	76 (88.37)	49 (87.50)
Yes	17 (11.97)	10 (11.63)	7 (12.50)
Hepatitis, n (%)				χ2=3.571	0.059
No	29 (20.42)	22 (25.58)	7 (12.50)
Yes	113 (79.58)	64 (74.42)	49 (87.50)
Number of TACEs in the first treatment cycle, M (Q1, Q3)	1.00 (1.00,1.00)	1.00 (1.00,1.00)	1.00 (1.00,1.00)	Z=−0.012	0.990
Number of RFAs in the first treatment cycle, M (Q1, Q3)	1.00 (1.00,1.00)	1.00 (1.00,2.00)	1.00 (1.00,1.00)	Z=−4.738	<0.001
AFP, ng/ml, M (Q1, Q3)	14.35 (4.40, 235.00)	12.60 (3.60, 170.30)	17.85 (4.90, 277.40)	Z=1.017	0.309
Tbil, μmol/L, M (Q1, Q3)	16.50 (10.90, 20.80)	16.05 (10.90, 22.06)	16.85 (11.70, 20.27)	Z=0.054	0.957
Albumin, g/L, Mean±SD	38.55±6.07	38.55±5.87	38.55±6.42	t=0.00	0.998
ALT, U/L, M (Q1, Q3)	25.36 (18.20, 39.69)	22.69 (15.40, 35.04)	37.25 (19.65, 51.22)	Z=3.465	<0.001
AST, U/L, M (Q1, Q3)	30.60 (22.85, 42.50)	27.70 (22.50, 40.50)	33.20 (23.60, 44.10)	Z=2.039	0.041
LDH, U/L, Mean±SD	189.82±41.75	189.73±41.59	189.96±42.37	t=−0.03	0.975
ALP, U/L, M (Q1, Q3)	91.50 (70.00, 126.00)	87.50 (70.00, 116.70)	100.95 (70.50, 139.95)	Z=1.062	0.288
Child-Pugh, n (%)				χ2=0.367	0.545
A	127 (89.44)	78 (90.70)	49 (87.50)
B	15 (10.56)	8 (9.30)	7 (12.50)
BCLC, n (%)				χ2=0.183	0.913
A	69 (48.59)	43 (50.00)	26 (46.43)
B	58 (40.85)	34 (39.53)	24 (42.86)
C	15 (10.56)	9 (10.47)	6 (10.71)
CNLC, n (%)				χ2=2.967	0.563
Ia	54 (38.03)	35 (40.70)	19 (33.93)
Ib	30 (21.13)	18 (20.93)	12 (21.43)
IIa	22 (15.49)	10 (11.63)	12 (21.43)
IIb	16 (11.27)	11 (12.79)	5 (8.93)
IIIa	20 (14.08)	12 (13.95)	8 (14.29)
The outcome, n (%)				χ2=48.964	<0.001
No progression or survival	86 (60.56)	72 (83.72)	14 (25.00)
Progression or death	56 (39.44)	14 (16.28)	42 (75.00)

PVTT – portal vein tumor thrombus; TACE – transarterial chemoembolization; RFA – radiofrequency ablation; AFP – alpha-fetoprotein; Tbil – total bilirubin; ALT – alanine transaminase; AST – aspartate transaminase; LDH – lactic dehydrogenase; ALP – alkaline phosphatase; BCLC – Barcelona Clinic Liver Cancer; CNLC – China liver cancer staging.

Development and Performance of the Decision Tree Model to Distinguish High-Risk and Low-Risk Patients

After LASSO regression analysis to screen out variables, we developed the decision tree model (Figure 1). A total of 142 samples were divided into 2 groups according to the number of RFA in the first treatment cycle ≤1.5. One group (samples=28) had reached the decision endpoint, and the other 114 samples were further divided into 95 samples and 19 samples according to Tbil ≤26.795. The dataset of 95 samples was divided into 2 groups according to ALT ≤14.55, 1 of which reached the decision endpoint (samples=14). The dataset of 19 samples continued to be classified according to ALT ≤39.12, and both datasets reached the decision endpoint (samples=12 and 7). As shown in Figure 1, the remaining samples were classified according to ALP level ≤152.45, AST ≤23.85, AST ≤28.95, AFP ≤1.95, and sex. Value=[] is the frequency of outcomes, the number on the left was progression-free and survival (low-risk group) and the number on the right was progressive disease or death (high-risk group) within 1 year. The cut-off of the decision tree model was 0.571, the area under the curve (AUC) and the positive predictive value (PPV) with 95% confidence interval (CI) was 0.846 (95% CI: 0.780–0.911) and 0.750 (95% CI: 0.637–0.863), respectively (Table 2).

Figure 1

Decision tree model to distinguish high-risk and low-risk patients. Patients were classified according to the indicated cut-off values of the variables. Value=[] is the frequency of outcomes, the number on the left is progression-free and survival (low-risk group), and the number on the right is progressive disease or death (high-risk group) within 1 year. The grid without judgment conditions has reached the endpoint of the decision path. Samples is the number of samples in the node. Num_RFA – number of RFA in the first treatment cycle; Tbil – total bilirubin; ALT – alanine transaminase; ALP – alkaline phosphatase; AST – aspartate transaminase; AFP – α-fetoprotein. (Python v. 3.7.4 sklearn package plot tree, Python Software Foundation, DE, USA).

Table 2

Predictive performance of decision tree model.

Cut-off	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	AUC (95% CI)	Accuracy (95% CI)
0.571	0.750 (0.637–0.863)	0.837 (0.759–0.915)	0.750 (0.637–0.863)	0.837 (0.759–0.915)	0.846 (0.780–0.911)	0.803 (0.737–0.868)

PPV – positive predictive value; NPV – negative predictive value; AUC – area under the curve; CI – confidence interval.

Comparison of the High-Risk and Low-Risk Groups

The comparison of low-risk and high-risk groups receiving combined TACE and RFA therapy is shown in Table 1. The ALT and AST levels were significantly higher in the high-risk group compared to the low-risk group (Table 1). There were significant differences in sex, age, tumor number, and number of RFA in the first treatment cycle between the high-risk and low-risk groups. Differences between the data before and after imputation were not statistically significant, indicating that our results were reliable (Supplementary Table 1). Figure 2 shows the cumulative incidence of progressive disease or death in the low-risk and high-risk groups. The risk of progressive disease or death in the high-risk group was 12.232 times (HR=12.232, [95% CI: 6.005–24.916]) that of the low-risk group after adjusting for the number of TACEs during the first treatment, LDH, tumor location, tumor size, PVTT, hepatitis, CP, BCLC, and CNLC (Table 3).

Figure 2

Kaplan-Meier curves for progressive disease or death according to patients in low-risk and high-risk groups (SAS v. 9.4, SAS Institute, Cary, North Carolina).

Table 3

Association of the high-risk group with progression or death in 1 year compared with the low-risk group.

	Model 1		Model 2		Model 3
	HR	P	HR	P	HR	P
High-risk group compared to low-risk group	6.70 (3.647–12.309)	<0.001	6.651 (3.620–12.221)	<0.001	12.232 (6.005–24.916)	<0.001

HR – hazard risk; CI – confidence interval. Model 1: did not adjust for confounders; Model 2: adjusted the number of TACE during the first treatment and LDH; Model 3: adjusted the number of TACE during the first treatment, LDH, tumor location, tumor size, PVTT, hepatitis, CP, BCLC, and CNLC.

Discussion

HCC is the most common pathological type of liver cancer. TACE and RFA are commonly used local treatment options. The combined use of these 2 can make up for their respective deficiencies and further exert their respective curative effects. Identifying the characteristics of high-risk groups from the patients undergoing combined TACE and RFA treatment and intervening in a timely manner is important to improve the quality of life and prognosis of HCC patients. Our decision tree model for dividing patients into low-risk and high-risk groups performs well. Compared with the low-risk group, the high-risk group was younger, had a higher proportion of males, had more tumors, received less RFA in the first treatment cycle, and had higher ALT and AST levels.

A retrospective study of HCC patients by Yuan et al [20] showed that the number of tumors was an independent factor affecting the overall survival of patients, which was consistent with the relevant literature [21]. Our results also showed that the high-risk group had more tumors than the low-risk group because the number of tumors can predict the degree of intrahepatic spread of HCC. The greater the liver tumor burden, the greater the size and extent of liver tumor infiltration [22].

The patients in the high-risk group were those with progressive disease or death, and we found these patients were younger than those in the low-risk group. The study by Gu et al showed that younger patients (<60 years old) were associated with recurrence of HCC, which was similar to our results [23]. Clinically, the pathological types are usually low in differentiation for young HCC patients, and poorly differentiated HCC is generally more malignant [24,25], so the prognosis after treatment is poor.

The results showed that the high-risk group received fewer RFA treatments in the first treatment cycle than the low-risk group. High-risk patients were younger, with lower tumor differentiation and higher malignancy. Clinically, these patients also had larger tumors and more obviously abnormal blood vessels. Therefore, TACE treatment is the main treatment, and sometimes multiple TACE treatments are required. If the embolization effect is obvious after repeated TACE treatment and the blood supply is reduced, RFA treatment can be performed on the area with blood supply [26-28]. RFA can regulate local tumor tissue to above 60°C, resulting in complete tumor degeneration and necrosis [29]. High temperature is the fundamental mechanism by which RFA to achieves tumor treatment [26]. If the embolization effect is not particularly satisfactory after each TACE, the TACE treatment is repeated. Therefore, high-risk patients received fewer RFA treatments.

In patients with chronic hepatitis and cirrhosis, the AST/ALT ratio was associated with progressive hepatic impairment [30]. As reported in previous studies, elevated AST levels were also found to be one of the prognostic factors for poor prognosis [31]. A retrospective paired case-control study conducted in a tertiary cancer center from January 2006 to December 2010 showed that ALT was an important prognostic factor in HCC patients treated with combined TACE and RFA [32]. Our study also found that ALT and AST levels were higher in the high-risk group than in the low-risk group. This may be because the combined therapy does not significantly damage liver function, and clinicians can monitor ALT and AST levels and perform appropriate interventions.

The main advantage lies in the risk stratification of HCC patients receiving TACE and RFA combined treatment by using the decision tree model and exploring the characteristics of patients under different risk stratification to provide a reference for clinical intervention. The present study has some limitations. The sample size of the enrolled cases in this study was small, and this was a retrospective single-center study. We did not collect data on clinical variables after the administration of combined TACE and RFA in HCC patients, so we could not demonstrate the treatment response. Large-scale, high-quality, prospective controlled trials are needed for further research in clinical practice.

Conclusions

To improve the prognosis and reduce the risk of progressive disease or death in HCC patients receiving combined TACE and RFA therapy, clinicians may need to consider the frequency of RFA administration during the first treatment and focus on the patients with higher ALT and AST levels.

Supplementary Table 1

Comparison between groups before and after data imputation.

Variables	After (n=142)	Before (n=142)	Statistics	P
AFP, ng/ml, M (Q1, Q3)	14.35 (4.40, 235.00)	14.35 (4.55, 203.80)	Z=0.365	0.715
Tbil, μmol/L, M (Q1, Q3)	16.50 (10.90, 20.80)	16.50 (10.90, 20.80)	Z=−0.045	0.964
Albumin, U/L, Mean±SD	38.55±6.07	38.47±6.03	t=0.10	0.919
ALT U/L, M (Q1, Q3)	25.36 (18.20, 39.69)	26.02 (18.30, 39.69)	Z=0.102	0.918
AST U/L, M (Q1, Q3)	30.60 (22.85, 42.50)	30.50 (22.85, 40.90)	Z=−0.088	0.930
LDH, U/L, Mean±SD	189.82±41.75	189.80±43.16	t=0.00	0.996
ALP, U/L, M (Q1, Q3)	91.50 (70.00, 126.00)	92.00 (71.00, 126.00)	Z=0.164	0.870

AFP – alpha-fetoprotein; Tbil – total bilirubin; ALT – alanine transaminase; AST – aspartate transaminase; LDH – lactic dehydrogenase; ALP – alkaline phosphatase.