Abstract
Background The selection criterion for hepatic resection(HR) in intermediate-stage(IM) hepatocellular carcinoma(HCC) is still controversial. This study aims to compare transarterial chemoembolisation (TACE) and HR in the range of predicted overall mortality(OM).
Methods In all, 946 consecutive patients with IM-HCC were categorised in HR and TACE group. We performed multivariable Cox regression model to predict OM in HR patients. To evaluate the HR impact on OM concerning baseline characteristics, we test the interaction between predicted OM risk and HR status. The cut-off values were determined by two-piece-wise linear regression model and decision curve analysis. Also, the inverse probability of treatment weight was performed to minimise potential differences as a sensitivity analysis.
Results Totally, 23.0% (n=225) of patients received HR. The 5-yr overall survival rate was higher in the HR group versus the TACE group (52.3% vs 22.8%; P<0.0001). In the HR group, five predictors (all<0.05) were selected to calculate the 5-yr OM risk. This model also used to predict the 5-yr OM-free survival rate. The line of HR and TACE was crossing with predicted OM risk at 100%. The benefit of HR versus TACE decreased progressively as predicted OM risk>55%. When OM risk >80%, HR was not significantly superior to TACE (HR:0.61;95%CI:0.31,1.21), and both HR and TACE did not increase net benefit.
Conclusions Hepatic resection was superior to transarterial chemoembolisation for intermediate-stage hepatocellular carcinoma at the 5-yr OM risk<80%.
1. Introduction
Hepatocellular carcinoma(HCC) is the fifth cause of death in China(1). According to the last BCLC staging system, transarterial chemoembolization(TACE) has been recommended as the first-line treatment for unresectable intermediate-stage HCC(2). In the clinical practice from Asia–Pacific region(3), intermediate eve advanced stage HCC are not the contraindication for surgical resection, including over three intrahepatic lessons, both lobes with tumors or satellite nodules. Although advantage of surgery over TACE in the resectable BCLC-B HCC patients has already been extensively described in the literature(4), it remaines an open issue which groups are more suitable for hepatic resection(HR) is still controversial.
Recently, a regret-based decision curve analysis (Regret-DCA) is performed to choose HR or TACE for IM-stage HCC(5). In this study, patients were grouped into low-risk (3yr mortality ≤ 35%), medium-risk (35%< 3yr mortality < 70%) and high-risk (3yr mortality ≥ 70%) groups by Regret-DCA. And only the patients in the low-risk gourp have the better outcome than who treat with TACE. However, more convenient biomarkers are urgent to choose suitable therapy. In this study, we use a predicted mortality risk-based decision analysis to compare the clinical outcome of hepatectomy and TACE in the IM-HCC.
2. Methods and patients
2.1 Patient selection
Clinical and biological data have been previously published in full(6). In this secondary analysis study, we mainly focused on the derivation cohort at Sun Yat-sen University Cancer Center. And the details of inclusion criteria were shown in Fig S1. In the derivation cohort, 946 patients were included into final analysis, with TACE (n=717, 73.2%) or surgical resection (n=225, 23.0%) as their first-line treatment.
The study protocol (2017-FXY-129) was approved by the Ethics Committee of SYSUCC and another three medical centers(6). Because this was a retrospective study, the informed consent was waived.
2.2 Definition and follow up
HR: surgical therapy for the lesions in hepatic segments or lobes. In the clinical, patients with well liver function and less tumour loading were usually suitable for HR.
TACE: chemoembolisation of the hepatic artery. According to the response to the TACE, the subsequent therapies include ablative therapies, surgical resection, TACE or targeted therapies.
Overall survival (OS): the time between the HR/TACE and death from any cause.
During the period of initial treatment for the first 2 years, patients were followed up for every 2 or 3 months if complete remission was achieved. The frequency gradually decreased to every 3 to 6 months after 2 years’ remission.
3.3 Statistical analyses
To compare differences of baseline characteristics between the HR and TACE groups, we compared categorical variables with the chi-square test and continuous variables by the Mann-Whitney test.
To test the OM-free survival between HR and TACE in the range of OM risk, we divide the statistical analyses into two main parts
Firstly, Kaplan-Meier methods were used to calculate OM-free survival rates for the HR and TACE cohorts. Also, we employed multivariable Cox regression (MVR) to develop the MVR model. In this study, the covariates (all P<0.05 between two groups) list in Table 1 were used to build the MVR model18, including the continuous covariates of age, PT and diameter of the main tumour, as well as the categorical covariates No. of intrahepatic lesions, both lobe with lesions. To assess discrimination of the model, the area under the receiver operating characteristic curve (AUC) was calculated. Besides, through decision curve analysis, we calculated the net benefit of the model and determined the cut-off value through two reference strategies (test none or test all).
Secondly, to evaluate the interaction between HR and OM risk, we used the method(7) as follow: (1)we used Model □ to test the association between OM and covariates in HR patients. From this model, OM-risk at 5 yr was predicted to establish a baseline OM risk for both HR and TACE cohort, (2) this predicted OM risk was added as a covariate to a second multivariate Cox model to calculate the predicted probability of survival at 5 yr, and (3) interaction between HR and OM risk was graphed using spline smoothing based on generalised additive model. Furthermore, a two-piece-wise linear regression and the recursive method was performed to calculate the inflexion point of the TACE line, and a log-likelihood ratio test was used to compare the one-line linear regression.
We performed the number needed to treat (NNT) analysis to explore how the NNT varies with predicted OM risk at 5yr. Stratified analysis was performed to examine the hazard ratio(vs TACE) for each different 20% proportion based on 5-yr predicted OM-risk. Meanwhile, the 5-yr observed survival probability of TACE cohort was estimated by the Kaplan-Meier curve. Then, we calculate NNT by the formulas of Altman and Andersen(8). Statistical analysis was performed using Empower (www.empowerstats.com, X&Y solutions, inc. Boston MA) and R software (version 3.4.3). P value < 0.05 considered significant.
3. Results
3.1 Descriptive characteristics
A total of 946 HCC patients were included into the derivation cohort, with a median age of 53.9 (SD, 12.3) years for TACE group(n=717), and 50.9(SD, 12.6) years for HR group(n=225). In the derivation cohort, patients with HR were younger, higher PT value, the shorter diameter of the main tumour, less frequently No. of intrahepatic lesions and both lobe with lesions (all P<0.05), which was shown in Table 1. The majority of the patient (825/942,87.6%) had hepatitis B virus (HBV) infection and the difference of HBV infection rate was not significant between HR and TACE group.
3.2 Survival analysis for the entire cohort
As was shown in Fig 1, the mOS is 18.5(95%CI:16.9, 20.3) month for TACE group versus 67.4(95%CI:46.7, 88.1) for HR group (p<0.0001). The median overall survival(mOS) for the entire cohort was 23.7 (95%CI:20.4, 27.2) month. There were 89 cases (89/942, 9.4%) still at risk at 5 yr. At 5 yr, OM-free survival rates was 30.6% (95%CI: 27.0%, 34.6%) for the entire cohort, and it was 52.3% (95% CI:44.9%, 61.0%) for HR group versus 22.8%(95%CI: 19.0%, 27.4%) for the TACE group (p < 0.0001).
In multivariable analysis focusing on the entire cohort (Table 2), age(HR,1.00; 95%CI: 0.99, 1.01), PT(HR,1.07; 95%CI:1.00, 1.14), diameter of main tumor(HR,1.15; 95%CI:1.13, 1.18), No. of intrahepatic lesions(HR,1.12; 95%CI:0.80, 1.57 for 3 vs 2; HR,1.59;95%CI:1.31, 1.93 for >3 vs 2, respectively), both lobe with lesions(HR,1.51; 95%CI:1.26, 1.80) were identified as the predictor for the MVR model. By decision curve analysis, we found that none could receive a net benefit from HR and TACE for OM risk >80%; maximal utility occurred at 55%(Fig S2).
Next, the predictors in MVR model were used to predict 5-yr OM risk and OM-free survival rate for the entire cohort. The predict OM risk was plotted against OM-free survival, and the lines for HR and TACE crossed at 100%(Fig 2). Consistent with the maximal utility point of DCA, the inflexion point of TACE line was calculated at 55%(p< 0.001, Table 3), indicating the benefit from HR decreased progressively as predicted OM risk >55%. Interestingly, the predicted survival rate of TACE line at 55% was 20% (95%CI:19%, 22%), which was in the interval of observed OM-free survival at 5yr for TACE group(22.78%, 95%CI: 18.96%, 27.37%).
Fig 3 showed the results of NNT analysis. After incorporating the missing data into the group of 40-50% (Table S1), hazard ratios were 0.40(95%CI: 0.26, 0.61), 0.41(95%CI: 0.28, 0.60), 0.40(95%CI:0.24, 0.68) and 0.61(95%CI:0.31,1.21) for each group. According to the formulas, NNT was not significantly available at the group of OM risk>80%.
4. Discussion
In this large-scale, real-world data, the 5-yr OM-free survival curve of HR and TACE were comparing in predicted OM risk for the range of 20%-100%. We found that the overall survival for HR was significantly better than their TACE counterparts, which was consistent with previous literature(4, 9). However, the net benefit from HR decreased progressively as predicted OM risk >55%. When OM risk>80%, HR was not significantly superior to TACE for the patients with BCLC-B HCC, and both HR and TACE did not increase net benefit.
Our study had several significant findings. When compared with TACE group, we found that the benefit from HR was nonlinearly associated with baseline 5-yr OM risk; maximal net benefit occurred at 55%. In 2015, Colombo et al(10) had come up with an assumption that intermediate-stage HCC patients could still be suitable for liver resection if the 5-yr survival rate reached 50%. Our findings primarily validated this hypothesis. In line with previous literatures(11, 12), we identified a subgroup with OM risk<80%, in which patients treated with HR had significantly better overall survival than TACE group. Based on the Bolondi’s sub-staging model(13), Wei et al(11), the postoperative 5-yr survival rate for patients in the BCLC B1-B3 stage was 49.5%, 33.7% and 12.9%, respectively. Nevertheless, only the patients with BCLC B1/B2 had an optimal long-term survival than TACE group. In another large-scale study(12), the benefit from liver resection was observed in the patients of BCLC-B1/B2 but not B3/B4.
Our study had some strengths, including providing the cut-off accuracy value to evaluate the tumor loading. Besides, in our study, the survival rates between HR and TACE were compared in the vast and continuous range, so that the exact cut-off values could be calculated. When evaluating the role of liver resection among patients with anatomically resectable tumor and well liver function, the randomized control trial was obviously against medical ethics. Therefore, the predicted mortality risk-based decision analysis for the real-world data may be a better choice.
Our study also had several limitations. Firstly, this was a retrospective cohort from real-world data, residual bias, and unmeasured confounders were unavoidable. Secondly, whether the liver resection was superior to TACE in western populations was still unknown. Thirdly, the percentage of resectable HCC patients in the TACE group with 5-yr OM risk< 80% were unclear. However, it was worthy of note that the potential unresectable HCC patients treated with TACE resulting from such errors would bias toward to the null and lead to an underestimation of the net benefit from liver resection versus TACE.
Data Availability
The raw data were freely obtained from the Dryad Digital Repository database (www.Datadryad.org; https://doi.org/10.5061/dryad.pd44k8r).
Ethical statement
The study protocol (2017-FXY-129) was approved by the Ethics Committee of SYSUCC and another three medical centers. Because this was a secondary analysis study, and the data were anonymous, the requirement for informed consent was waived.
Author Contributions
Conception and design: Siyu Chen, Xinyu Hu; Administrative support: Lijun Zhu; Provision of study materials or patients: Shan Lin, Lijun Zhu;Collection and assembly of data: Siyu Chen, Xinyu Hu; Data analysis and interpretation: Siyu Chen, Xinyu Hu; Manuscript writing: All authors; Final approval of manuscript: All authors
Supplemental materials
Table S1. The stratified analysis of predicted OM risk at 5yr before/after incorporating the missing data. Model□ showed the missing data were regarded as a independent group. Model □ showed the missing data were incorporated into the group of 40-50%.
Fig S1. Inclusion criteria of hepatocellular carcinoma patients. SYSUCC=Sun Yat-sen University Cancer Center. HCC=hepatocellular carcinoma.
Fig S2. Net benefit curves for the multivariable Cox regression model. The y-axis measures standardized net benefit.The maximal net benefit occurred at 0.55.
Acknowledgments
We gratefully thank the Empower U team of the Department of Epidemiology and Biostatistics, X&Y solutions Inc. in Boston, for their contribution to the statistical support.
Footnotes
Source of Funding:Dr.Zhu received financial support from the Natural Science Foundation of 900th hospital of PLA (Nos 2016Q05 and 2017Q02 and 2017L17 and 2016L03 and 2017J03).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form. The authors have no conflicts of interest to declare
author affiliations updated