Predicting survival in patients with hepatocellular carcinoma treated by transarterial chemoembolisation


Prof. C. Cammà, Sezione di Gastroenterologia, Dipartimento Biomedico di Medicina Interna e Specialistica, University of Palermo, Piazza delle Cliniche 2, 90127 Palermo, Italy.


Aliment Pharmacol Ther 2011; 34: 196–204


Background  Transarterial chemoembolisation (TACE) is first-line treatment in unresectable hepatocellular carcinoma (HCC) and rescue treatment after failure of radical treatments in early stage HCC. Prognostic tools for HCC using time-fixed Cox models may be unreliable in patients treated with TACE because time-varying predictors interact.

Aim  To explore time-dependent variables as survival predictors in patients with HCC receiving TACE as first-line or second-line treatment.

Methods  Eighty four consecutive patients with HCC (mean age 68; male gender 62%; Child-Pugh class: A n = 73, B n = 11; Barcelona Clinic Liver Cancer class: A n = 44, B n = 24, C n = 16) treated with TACE were enrolled. Clinical, laboratory and radiological follow-up data were collected from the time of first treatment. Time-fixed and time-dependent Cox analyses were done.

Results  Overall survival rates were 89.6% (95% CI 82.5–97.2) at 12 months, 58.8% (95% CI 46.2–74.9) at 24, 35.4% (95% CI 22.3–56.1) at 36 and 17.2% (95% CI 7.0–41.7) at 48 months. Performance status (P < 0.001), number of nodules (P < 0.016) and prior therapy (P = 0.017) were the only variables strongly linked to survival by time-fixed Cox model. Performance status (P < 0.001), prior therapy (P = 0.005), number of treatments (P = 0.013), complete response after TACE (P = 0.005) and bilirubin level (P < 0.001) were associated with survival using a time-dependent Cox model.

Conclusions  Survival after TACE is influenced most by performance status, complete response and bilirubin. Compared with the time-fixed models, a time-dependent Cox model has the potential to estimate a more precise prognosis in HCC patients treated with TACE.


Transarterial chemoembolisation (TACE) is the most widely used first-line treatment in the Western world and Asia for patients with unresectable hepatocellular carcinoma (HCC). TACE is considered the standard of care in these patients on the basis of two published meta-analyses that cleared demonstrated 2-year overall survival benefit.1, 2 There is sufficient evidence from randomised controlled trials (RCTs) on the type of patients who are candidates for TACE. In addition, the American Association for the Study of Liver Diseases (AASLD) Practice Guideline for the management of HCC indicates TACE in nonsurgical patients at intermediate Barcelona Clinic Liver Cancer (BCLC) B stage.3 Nonetheless, application of TACE in real clinical practice remains a matter of debate.

The selection criteria used to identify the candidates for TACE vary nationally and internationally.2 Optimal candidates are patients with excellent liver functional reserve who are not presented with cancer-related symptoms or vascular invasion.3, 4 However, intermediate BCLC B patients may have ascites, which is strongly linked to poor survival in this stage.5, 6 Moreover, TACE can be performed in early-stage (BCLC A) patients for whom percutaneous radiofrequency thermal ablation (RFTA) is not feasible because of tumour location (proximity to gall-bladder, biliary tree, or blood vessel) or on whom surgery cannot be performed because of comorbidities.7

Despite the many RCTs that have been done to identify the optimal chemoembolisation procedure, lack of standardisation affects some aspects of treatment, including embolisation technique and treatment schedules.8, 9 A relevant clinical question is whether patients should receive repeated courses of TACE at fixed intervals until the planned numbers of courses is reached or until a complete radiological response is achieved. There is no evidence that complete radiological response after TACE is a surrogate end-point strongly linked to overall survival.

A major problem in assessing overall survival of patients with HCC secondary to cirrhosis arises from a lack of accurate models able to predict outcome in the individual patient.3, 10–12 Two factors may contribute to a less than satisfactory performance.

First, all previous prognostic models were general, not specialised, models for evaluating patient outcomes within specific populations or allocation treatment groups (e.g. liver transplantation, percutaneous ablation, TACE and systemic therapies).

Second, all the previous prognostic models assumed that the variables noted at one single time point for each patient were sufficient for predicting survival. In HCC, as in most chronic diseases, the clinical situation changes with time, particularly after a treatment. Conceivably, the weight of prognostic indicators changes in different phases of the disease, and estimates of prognosis could improve if such time-dependent changes were taken into account.

The aims of this prospective cohort study of HCC patients treated with TACE as first-line or second-line treatment were:

  • (i) to assess the effectiveness of TACE in HCC patients;
  • (ii) select the optimal candidate for TACE by identification of predictors of overall survival; and
  • (iii) improve the accuracy of mortality risk estimates with a new prognostic model that accounts for changes during the course of the disease.

Materials and methods


From January 2004 to May 2009, 84 consecutive HCC patients were treated with TACE as first-line or second-line treatment at our institution. Follow-up was censored on 31 October 2009. HCC was diagnosed and staged, respectively, according to AASLD criteria and BCLC schedule.3

Patients with early tumours (single tumours measuring less than 5 cm, or three nodules measuring less than 3 cm) were considered for curative therapies. Resection was indicated for patients with single tumours, absence of portal hypertension and normal bilirubin concentrations. Patients with portal hypertension or abnormal bilirubin concentrations or three nodules of less than 3 cm in diameter were considered for transplantation. Percutaneous treatment was performed when surgery was precluded.

TACE was performed in patients:

  • (i) who were not suitable for curative treatments because of locally advanced HCC (tumour size >5 cm, or multifocal disease);
  • (ii) in those for whom percutaneous treatments were precluded due to position (pericholecystic, periportal, subfrenic or subcapsular lesions); and
  • (iii) in those for whom prior curative treatments had failed.

Exclusion criteria were portal vein thrombosis, impaired liver function (Child-Pugh class ≥B8), widespread cancer (defined as involving more of the 50% of the liver, or with extra hepatic metastasis), hepatic encephalopathy and ascites.

Extra-hepatic disease was assessed with multidetector multiphasic CT and chest radiography. Bone metastases were sought using scintigraphy if clinically suspected.

TACE technique

Each patient underwent standard hepatic angiography using the Seldinger procedure. The catheter tip was advanced as near as possible to the feeding artery. An emulsion of anticancer agent (doxorubicin) and lipiodol, followed by gelatin sponge particles, was carefully injected during the X-ray monitoring. The dose of anticancer agent emulsion and lipiodol, as well as the pieces of embolic materials used for TACE, were based on tumour size and extension of the lesions. TACE was done at baseline and, when there was evidence of any tumour persistence, every 2 months thereafter. It was withheld or discontinued whenever vascular contraindications, poor hepatic function, severe adverse effects or progressive disease developed. One month after every TACE, a multiphasic CT scan was done.

Outcomes and follow-up

The primary outcome was overall survival. Follow-up time was defined as the number of months from first TACE to orthotopic liver transplantation (OLT), other treatments, last contact with the patient or death.

Response rate magnitude was defined according to the European Association for the Study of the Liver (EASL) criteria.13 These are addressed by recent guidelines.14, 15 EASL response criteria are defined as follows:

(i) complete response, defined as absence of any enhancing tissue;

(ii) partial response (PR), defined as >50% decrease in enhancing tissue; and

(iii) stable disease, defined as <50% decrease in enhancing tissue.

Progressive disease is defined as any increase in enhancement of the treated tumour.

As we used an on demand protocol, assessment of response was not done at a fixed time, but after the last performed TACE. The follow-up protocol included clinical assessment by physical examination, ultrasound scan and biochemistry every 3 months and using multiphasic CT scan every 6 months. In our study, TACE-related morbidity was defined as any complication within 2 weeks of each session of TACE. TACE-related mortality was defined as death from a complication within 2 weeks of each session of TACE. Decompensation of liver disease was defined as appearance of ascites, jaundice and/or hepatic encephalopathy during follow-up.

Statistical evaluation.  The Kaplan–Meier model was used to estimate survival. Differences in the survival rate were assessed by log-rank testing. Potential prognostic variables were evaluated as predictors of survival both in time-fixed and time-dependent Cox models.16, 17 In the time-fixed model, only the initial records at the time of first TACE were applied (Table 1). The time-dependent model used the repeated measurements of the potential prognostic variables, with the addition of treatment response, time to progression (TTP) and number of treatments, during follow-up.

Table 1.   Demographic, laboratory, clinical and tumour staging characteristics of 84 patients with HCC in compensated cirrhosis treated with TACE
  1. IU, International Units; AFP, alpha fetoprotein; HCV, hepatitis C virus; HBV, hepatitis B virus; BCLC, the Barcelona Clinic Liver Cancer; MELD, Model for End-Stage Liver Disease.

  2. Values are mean ± s.d.

  3. * Eastern Cooperative Oncology Group – performance status.

Age – years68 ± 7
Male –n (%)52 (62)
Aetiology of cirrhosis –n (%)
 Anti-HCV positivity69 (82)
 HBsAg positivity11 (13)
 Anti-HCV positivity plus Alcohol abuse2 (2.5)
 NAFLD2 (2.5)
Performance status*–n (%)
 068 (81)
 116 (19)
Hepatic encephalopathy –n (%)
 None84 (100)
Ascites –n (%)
 None84 (100)
Albumin – g/dL3.6 ± 0.5
International normalised ratio1.10 ± 0.12
Total bilirubin – mg/dL1.3 ± 0.7
Child-Pugh Score5.8 ± 0.9
Classes –n (%)
 A73 (87)
 B11 (13)
Platelet × 103/mmc100 ± 50
Creatinine – mg/dL0.8 ± 0.3
MELD Score9.0 ± 1.9
Oesophageal varices –n (%)
 None15 (17.9)
 F132 (38)
 F230 (35.7)
 F37 (8.4)
Portal vein thrombosis –n (%)
 None84 (100)
AFP, median (range) (ng/mL)20 (3 – 2.000)
Number of nodules (%)
 158 (69)
 215 (18)
 36 (7)
 ≥45 (6)
Maximum tumour diameter (cm)3.3 ± 1.6
Number meeting Milan criteria –n (%)
 In60 (71)
 Out24 (29)
BCLC –n (%)
 A44 (52)
 B24 (29)
 C16 (19)
Prior therapy –n (%)
 None61 (73)
 Resection2 (2.3)
 Radiofrequency ablation19 (22.4)
 Percutaneous ethanol injection2 (2.3)

All variables studied at univariate analyses were entered into the multivariate analyses. To avoid the effect of co-linearity with performance status, BCLC score was not included in the same multivariate model. Variables with a P value of <0.10 at univariate analysis were included in the final multivariate models. A Cox model was used to identify prognostic factors for mortality in a multiple regression analysis using time-fixed and time-dependent analyses. All P values were two-tailed, and all confidence intervals (CIs) were 95%.

Time-fixed and time-dependent model discrimination were evaluated at 6, 12 and 24 months by the area under incident/dynamic receiver operating characteristic curves (AUROC) determined by the Heagerty and Zheng method.18 At every considered time, AUROCs were tested using a Wilcoxon rank sum test with continuity correction.19, 20

To have a broader view of the differences between the AUROCs, an Integrated Area Under the Curve (IAUROC) calculation was made.

All multivariate analyses were done with PROC PHREG in sas version 8.1 (SAS Institute, Inc., Cary, NC, USA). Model discrimination analyses were carried out with the R Statistical Computing, version 2.10 (R Foundation for Statistical Computing, Vienna, Austria).


Patient features at baseline

The study population consisted of 84 patients with HCC secondary to cirrhosis of different aetiologies. Of these 84 patients, 61 received TACE as first-line treatment, whereas 23 received it as second-line treatment. The demographical, clinical and tumour staging features of the 84 patients are given in Table 1. A single lesion was observed in 69%, two or three lesions in 25%, and more than three lesions in 6% of patients (Table 1).

At presentation 44 of the 84 patients (52%) were BCLC early (A) stage, 24 were BCLC intermediate (B) stage (29%) and 16 were BCLC advanced (C) stage (19%). Patients at BCLC C stage were classified in this stage because they had symptomatic disease with PS 1.


Three of the 84 patients (4%) were lost at follow-up. During follow-up, 34 patients died (Table 2), 12 patients (14%) were withdrawn for other treatments, whereas 26 (31%) were alive at the end of the study (Table 2). Among the 34 deaths, only three patients (4%) died cancer-free (i.e. without local recurrence, new lesions or distant metastases). Median overall survival after first TACE was 30.5 months [95% confidence interval (CI) 24.0–38.8] for the entire group. The overall survival rates (Figure 1) were 89.6% (95% CI 82.5–97.2); 58.8% (95% CI 46.2–74.9); 35.4% (95% CI 22.3–56.1); 17.2% (95% CI 7.0–41.7); and 5.7% (95% CI 0.9–35.7) at 12, 24, 36, 48 and 60 months, respectively.

Table 2.   Follow-up of 84 patients with HCC in compensated cirrhosis treated with TACE
  1. TACE, transarterial chemoembolisation; OLT, orthotopic liver transplantation; PEI, percutaneous ethanol injection; RFTA, radiofrequency thermal ablation; CI, confidence interval.

TACE-related mortality –n (%)0 (0)
TACE-related morbidity –n (%)
 Abdominal pain15 (18)
 Fever12 (14)
 Vomiting8 (9)
 Acute cholecystitis1 (1)
Response after TACE –n (%)
 Complete response30 (36)
 Partial response34 (40)
 Stable disease19 (23)
 Progressive disease1 (1)
Patients withdrawn for other treatments –n (%)
 OLT2 (2)
 PEI2 (2)
 RFTA4 (5)
 Sorafenib4 (5)
Time to progression – months (95% CI)9 (4–18)
Disease-free survival, median – months (95% CI)10 (9–12)
Death –n (%)34 (40)
Overall survival median – months (95% CI)30.5 (24.0–38.8)
Decompensation of liver disease –n (%)28 (82)
Portal vein thrombosis –n (%)14 (16.7)
Figure 1.

 Probability of overall survival according to Barcelona Clinic Liver Cancer in 84 patients with HCC in compensated cirrhosis treated with TACE.

Table 2 lists TTP and response rate according to EASL criteria.11 Response rate after TACE was 76% (complete response, 30%; partial response 34%), and the safety of TACE was found to be acceptable. No patient death was treatment-related. The most common symptoms were development of transient abdominal pain (18%), fever (14%), and vomiting (9%). The only major complication was acute cholecystitis (1%) (Table 2).

Analysis of factors affecting the survival of patients

Cox regression analysis by time-fixed model showed that performance status (HR 10.9; 95% CI: 4.4–27.1; P < 0.001), multiple nodules (HR 3.7; 95% CI: 1.2–8.6; P = 0.016) and no prior therapy (HR 3.3; 95% CI: 1.3–10.7; P = 0.017) were independent risk factors for mortality (Table 3).

Table 3.   Multivariate Cox regression models for predicting overall survival in patients with HCC in compensated cirrhosis treated with TACE
Variable (Code)HR95% CIP-value
  1. HR, hazard ratio; CI, confidence interval; TACE, transarterial chemoembolisation.

  2. * Eastern Cooperative Oncology Group – performance status.

Time-fixed approach
 Performance status* (0/1)10.94.4 – 27.1<0.001
 No prior therapy3.31.3 – 10.70.017
 Number of nodules (≥3)3.71.2 – 8.60.016
Time-dependent approach
 Performance status* (0/1)7.903.63 – 17.23<0.001
 No prior therapy6.31.77 – 22.630.005
 Number of treatments  (Continuous)0.50.31 – 0.870.013
 Complete response  after TACE (yes/no)0.10.03 – 0.520.005
 Log – bilirubin (Continuous)4.22.18 – 7.89<0.001

Cox regression analysis by time-dependent model showed that performance status (HR 7.9; 95% CI: 3.63–17.2; P < 0.001); no prior therapy (HR 6.3; 95% CI: 1.8–22.67; P = 0.005); multiple treatments (HR 0.5; 95% CI: 0.3–0.9; P = 0.013); complete response (HR 0.1; 95% CI: 0.03–52; P = 0.005); and elevated bilirubin (HR 4.2; 95% CI: 2.2–7.9; P < 0.001) were independent risk factors for mortality (Table 3).

When relapsing bilirubin levels as a continuous variable with bilirubin levels as categorical variable (cut-off ≤3 mg/dL) were measured, we obtained similar results. The discriminating ability of time-fixed and time-dependent models in predicting overall survival was evaluated by AUROC values at 6 months [0.901 for the time-dependent model, and 0.848 for the time-fixed model (Figure 2a)]; at 12 months [0.903 for the time-dependent model, and 0.817 for the time-fixed model (Figure 2b)]; and at 24 months [0.908 for the time-dependent model, and 841 for the time-fixed model (Figure 2c)]. AUROC of the time-dependent model was significantly higher than that of the time-fixed model at all time points [the time-dependent vs. time-fixed models two-sided P-values were <0.001, <0.001 and =0.01 at 6, 12 and 24 months, respectively (Figure 2a,b,c)].

Figure 2.

 Receiver Operating Characteristic (ROC) curves and area under the ROC Curve for Time-fixed and Time-dependent Cox regression models At 6 (a), 12 (b) and 24 months (c). (d) Comparison between the Time-dependent and Time-fixed Cox models in terms of integrated AUROCs.

The Integrated AUC (IAUC) of the time-dependent model was significantly higher than that of the time-fixed model (time-dependent model vs. time-fixed model, P-value <0.0001), demonstrating a better discriminating capacity of the time-dependent model (Figure 2d).

The estimated probability of survival in the four hypothetical patients treated with TACE who achieved complete response, according to factors significantly predicting mortality by the time-dependent model (i.e. performance status and bilirubin levels) are shown in Figure 3. The 3-year probability of overall survival for a patient with complete response, stable performance status 0 and normal bilirubin levels on treatment is 82.5%, and that of a patient with complete response without stable performance status and abnormal bilirubin levels is 0.5%.

Figure 3.

 Estimated probability of overall survival for different patients with HCC treated with TACE, according to variability in predicting factors by Time-dependent Cox regression model. Patients with complete response and (1) Stable performance status and bilirubin levels; (2) Stable performance status and impaired bilirubin levels; (3) Impaired performance status and stable bilirubin levels; (4) Impaired performance status and bilirubin levels.


Hepatocellular carcinoma secondary to cirrhosis is a complex and heterogeneous disease with wide variations during its clinical course.5, 21 None of the current general prognostic systems has provided confident prediction of survival in individual patients,10, 12 and so we urgently need specialised models to evaluate patient outcomes within specific allocation treatment groups.22

Another important issue is the lack of prognostic tools that are able to adequately express the complexity of the interactions during follow-up between tumour factors and degree of liver failure. As a result, we need models that account for changes during the course of the disease.

In this study, which is to our knowledge the first of its kind, the predictive accuracy of a time-dependent Cox model for overall survival prediction in patients with unresectable HCC and compensated cirrhosis treated with TACE was prospectively compared with that of the most widely-used time-fixed models. The performance of the time-dependent model was significantly better than that of the time-fixed model, and the overall predictive ability of this model, including simple and easy clinical and radiological predictors, was excellent. Moreover, it was uniform over the three time points considered: 6, 12 and 24 months.

This study shows that the key prognostic factors of HCC treated with TACE change over time, and that they can be combined into a specialised model that accurately estimates the risk of mortality at different time points during the post-treatment disease course.

From a practical point of view, the excellent prediction of survival achieved by our model could be useful for:

  • (i) controlling for confounding factors in observational studies;
  • (ii) calculating the sample size and stratifying patients in RCTs; and
  • (iii) assessing treatment effect size in order to formulate therapeutic strategies.

Our study also shows that patients with good performance status may benefit from TACE treatment both in the time-fixed and time-dependent models. The finding that a cancer-related variable, such as the number of tumour lesions assessed at baseline, is an independent predictor for better survival at multivariate analysis is in agreement with the results of a previous study by Takayasu.23 These two parameters, both included in the BCLC classification, may explain why this staging system provides accurate information on prognosis in the setting of HCC.24

An important finding in the present study was the significantly increased risk of death indicated by the time-dependent model in patients with elevating bilirubin levels during follow-up. Therefore, the results of TACE in terms of survival should be carefully evaluated because its indisputable antitumor effects are offset in practice by deterioration in liver function. It is conceivable that a benefit with chemoembolisation could be had only with the sort of careful selection that excludes patients with high risk of progressive liver failure after TACE.25 That the median survival observed in our study (30 months) was good was likely due to an accurate selection of patients.

As previously reported in the setting of treatment with RFTA,6 percutaneous ethanol injection (PEI),26 and, more recently, TACE,27 it would seem, from the present study, that overall survival depends strictly on complete radiological response. So, complete response is a relevant surrogate end-point that should be planned and carefully assessed by multiphasic CT scan in all patients who undergo TACE.

A general consensus on the optimal schedule of TACE treatment is yet to be reached.6 In various studies, treatment was repeated at fixed intervals based on a pre-established number of courses, or repeated on-demand based on radiological response. Our study suggests the possibility that repeated courses of treatment are more efficient than single treatment, and that a complete radiological response should be achieved because it is a relevant surrogate end-point. The possibility of achieving this surrogate end point with TACE should be carefully assessed before determining treatment failure after TACE, and then switching to sorafenib.

What are the implications of these results for current practice? In accordance with current AASLD guidelines,3 we confirmed that mortality risk estimates should be made at baseline with the BCLC staging system. Our study provides further evidence that the evaluation of other simple parameters during the course of treatment, namely performance status, radiological response and bilirubin levels by a time-dependent model can reliably aid in determining whether further sessions of TACE are needed. In the subgroup of patients with the most favourable on-treatment predictors, the probability of 3-year overall survival was 82%. In patients with complete response, but with coexistence of the most unfavourable on-treatment predictors (worsening of the performance status and increasing of bilirubin levels), the prognosis was very poor. Accordingly, TACE should be repeated until achieving complete necrosis in patients with a good and stable performance status and with relatively preserved liver function (bilirubin level <3 mg/dL).

Our analysis, specifically designed to identify prognostic factors of survival, shows that patients who received TACE as first-line treatment had a higher risk of death than those who received TACE as second-line treatment. The worse survival observed in the first group could be explained by the fact that the first group had a greater number of patients in a more advanced stage than did the second group. Moreover, in our prospective cohort study, 44 of 84 patients (52%) were in BCLC class A, and received TACE. Among these, more-curative treatment failed in 23 patients, who then received TACE as second-line therapy. The remaining 21 could not receive curative treatment because of tumour location and/or comorbidities. It should be recognised that not all patients in real clinical practice defined by each stage of BCLC are ultimately candidates for the suggested treatment modality.

Moreover, the fact that multiple treatments appeared to be a predictor of good outcome may reflect a selection bias due to which patients receiving multiple treatments were those that were healthy enough to tolerate multiple treatments over a long period.

Another weakness of our prognostic model is the lack of data on molecular factors, such as gene expression profiling, which can have some impact on patient’s outcome.28, 29 The predictive ability of our model, including simple and easy clinical and radiological predictors was excellent, underscoring the applicability of our results to new populations and settings, particularly in real clinical practice, where complex and expensive tests are not available.

Finally, we should be particularly concerned about the small number of patients included in this single centre study compared with other prospective, multicenter studies.

Further large-scale prospective studies may prove useful in substantiating the benefit of this new approach.

The available evidence is sufficient to conclude that:

  • (i) the time-dependent Cox model better predicts overall survival than the time-fixed model;
  • (ii) patients with a good performance status and with a low number of nodules at baseline, are good candidates for TACE;
  • (iii) a complete radiological response after TACE significantly increases overall survival, and should therefore be considered a surrogate endpoint of treatment; repeated courses of treatment are more efficient than a single treatment; and
  • (iv) bilirubin elevation during follow-up significantly increases the risk of death; therefore, treatment should be discontinued when bilirubin exceeds the value of 3 mg%.


This article is dedicated to the memory of Professor Luigi Sandonato. The authors thank Professor Jordi Bruix for his many suggestions. The authors thank Warren Blumberg for his forbearance in editing the manuscript. Declaration of personal and funding interests: None.