Potential conflict of interest: Nothing to report.
The presence of cirrhosis increases the potential risk of hemorrhage for patients with hepatocellular carcinoma (HCC). We evaluated the relative risk for hemorrhage in patients with HCC treated with antiangiogenic agents. We performed a systematic review and meta-analysis of antiangiogenic studies in HCC from 1995 to 2011. For nonrandomized studies we compared bleeding risk with other HCC single-arm studies that did not include an antiangiogenic agent. To separate disease-specific factors we also performed a comparison analysis with renal cell cancer (RCC)) studies that evaluated sorafenib. Sorafenib was associated with increased bleeding risk compared to control for all grade bleeding events (odds ratio [OR] 1.77; 95% confidence interval [CI] 1.04, 3.0) but not grade 3-5 events in both HCC and RCC (OR 1.46; 95% CI 0.9, 2.36; P = 0.45). When comparing the risk of bleeding in single-arm phase 2 studies evaluating antiangiogenic agents, this risk for all events (OR 4.34; 95% CI 2.16, 8.73) was increased compared to control. Conclusion: This analysis of both randomized and nonrandomized studies evaluating an antiangiogenic agent in HCC showed that whereas the use of sorafenib was associated with an increased risk of bleeding in HCC, this was primarily for lower-grade events and similar in magnitude to the risk encountered in RCC. (HEPATOLOGY 2013)
Until recently the role of systemic therapy in the management of hepatocellular carcinoma (HCC) was minimal. This changed with the publication of the landmark SHARP study in 2008, which resulted in sorafenib becoming the standard of care option for disease that is not amenable to surgery, ablation, or chemoembolization.1 Although it is true that the median survival advantage in this study was 3 months, its major importance arguably lay in the momentum that it gave to the field, and in particular to the development of so-called “antiangiogenic” therapies in HCC. However, antiangiogenic therapies carry their own particular risk profile—including bleeding, hypertension, proteinuria, and thrombotic events—and this profile has been further and better defined in the time since the first major study demonstrated proof of principle for their efficacy.2
In any HCC clinical trial the majority of patients will have underlying cirrhosis and this serves as an additional comorbidity that must be accounted for in the eligibility criteria and risk assessment. It also increases the baseline risk for a patient entering a study, with a greater potential for overlap between the cirrhosis-related risk and the toxicities of the agent under study. Of particular concern is the risk of bleeding in this patient population, who frequently suffer from portal hypertension and thrombocytopenia. However, there are no standardized eligibility criteria across HCC studies—as regards, for example, acceptable platelet count and coagulation parameters or mandated endoscopy to detect varices—to safeguard against this added risk of bleeding while at the same time taking into account the fact that HCC patients have baseline parameters that would ordinarily be exclusionary.
We sought to investigate fully the incidence and relative risk of bleeding events in patients with HCC who have been treated with an antiangiogenic agent, mainly sorafenib, as part of a clinical trial. Our major aim was to ascertain whether in fact the bleeding risk is increased in this patient population being treated with this class of drug. Because the majority of randomized studies in HCC have evaluated sorafenib, the greater part of our analysis pertained to this drug. To separate disease-specific factors from potential drug class effect we compared the risk of bleeding in HCC studies with that of randomized studies also evaluating sorafenib in renal carcinoma (RCC). We also set out to describe the considerable heterogeneity that exists with regard to the eligibility criteria for study entry in HCC.
Patients and Methods
Selection of Studies.
We performed a systematic review of published studies evaluating an antiangiogenic agent in patients with HCC. The term “antiangiogenic” was defined as any therapy whose putative mode of action was either wholly or partly directed against the tumor vasculature. To identify relevant clinical trials we conducted a PubMed search of citations from January 1995 to December 31, 2011. The search terms employed in our literature search included: “hepatocellular carcinoma,” “antiangiogenic,” “sorafenib,” “sunitinib,” and “bevacizumab.” For the randomized studies we used the nontreatment groups as control and for the nonrandomized single-arm phase 2 studies, which accounted for the majority of the studies, we compared bleeding risk with other HCC single-arm studies not including an antiangiogenic agent. To separate disease-specific effects we also performed a comparison analysis with RCC studies that evaluated sorafenib. We confined our analysis to prospective studies that have been published in article form.
We analyzed studies that met the following criteria: phase 1, 2, or 3 trials in HCC; phase 3 studies evaluating sorafenib in RCC; participants assigned to treatment with an agent whose mechanism of action was known to be wholly or partially antiangiogenic; adequate safety data available for bleeding events. For every study we extracted the following information: author name; year of publication; number of enrolled patients; treatment; eligibility criteria regarding platelet count, coagulation, hepatic function, Child-Pugh status; endoscopic requirements. Although we sought to evaluate cross-study variability in entry criteria we did include studies where the eligibility criteria were incomplete, as this was not the primary aim of our analysis. The occurrence of hemorrhagic events of any grade was recorded. We assessed and recorded adverse events according to the National Cancer Institute's common toxicity criteria (v. 2.0 or 3.0), which were used by all of the clinical trials.
To calculate incidence we extracted from the safety profile the number of bleeding events (all grade and grade 3-5) and the number of patients in the study. For every study we derived the proportion (and 95% confidence interval [CI]) of patients with adverse outcomes. For studies which contained a control arm the number of events was entered for both arms and the Mantel-Haenszel method used to calculate an odds ratio (OR) and 95% CI. These ORs were plotted in a forest plot where they were assigned a weight, based on sample size and variance, and pooled for an overall effect estimate of antiangiogenesis therapy on bleeding events. Analysis (using the inverse variance method, alpha of 0.05) and forest plots were generated using R statistical software and Review Manager.
CTCAE, Common Terminology Criteria for Adverse Events; HCC, hepatocellular carcinoma; RCC, renal cell cancer; VEGF, vascular endothelial growth factor.
To quantify the risk of hemorrhage associated with the use of antiangiogenic agents in HCC we performed an analysis of (1) randomized studies involving sorafenib in both HCC and RCC (to separate disease-specific effects) and (2) nonrandomized studies in HCC involving heterogeneous antiangiogenic therapies, which we compared with similar studies not involving antiangiogenic therapy. The randomized studies are shown in Table 1, whereas the nonrandomized studies involving both angiogenic and nonangiogenic therapies are shown in Tables 2 and 3, respectively. Six randomized studies involving a total of N = 2,464 patients and which compared sorafenib to control in HCC and RCC were identified. Of the HCC patients, 50% were enrolled in either the SHARP study or the Asia-Pacific study. Among the included set of HCC single-arm studies, there were 19 treated with antiangiogenic therapy and 21 treated with non-antiangiogenic therapy. Eleven of the studies involved sorafenib, four employed bevacizumab, and three studies evaluated sunitinib. The remaining studies involved other agents that are reported to have antiangiogenic effects (brivanib, lenalidomide, TSU68, linifanib, ramicirumab).
Table 1. Randomized Sorafenib Studies in RCC and HCC
Sorafenib+Doxorubicin vs. Placebo+Doxorubicin
A randomized controlled study of BAY43-9006 in combination with doxorubicin versus doxorubicin in patients with advanced hepatocellular carcinoma.
Sorafenib vs. Placebo
A randomized, double-blinded, placebo-controlled study of sorafenib in patients with advanced hepatocellular carcinoma.
Sorafenib vs. Placebo
Phase III randomized, double-blind, placebo-controlled trial of sorafenib in patients with advanced clear-cell renal cell carcinoma (RCC)
Sorafenib vs. Interferon
A randomized, open-label, multi-centre phase II study of BAY43-9006 (sorafenib) versus standard treatment with interferon alpha-2a in patients with unresectable and/or metastatic renal cell carcinoma
Sorafenib vs. Placebo
Phase III study of BAY43-9006 in patients with advanced hepatocellular carcinoma (HCC) treated after transcatheter arterial chemoembolization (TACE).
Sorafenib vs. Placebo
A Phase III randomized, placebo-controlled study of sorafenib in patients with advanced hepatocellular carcinoma.
Table 2. Eligibility Criteria and Bleeding Experience in HCC Studies Involving Antiangiogenic Therapy
Prior Bleeding or Anticoagulation Restriction
Child-Pugh (CP) Status/ Liver Function or ALT/AST
Llovet et al. (1)
Sorafenib vs. Placebo
CP A albumin, ≥2.8 bilirubin, ≤3 ALT/AST ≤5x ULN2
No stated restriction
Sorafenib versus placebo: 9% vs. 13% serious hemorrhagic event; 2% vs. 4% variceal bleeding; 0% vs. 2% peritoneal hem; 1% vs. 2% upper GI hem
Abou-alfa et al. (20)
Sorafenib + doxorubicin vs. doxorubicin
CP A bilirubin ≤3 mg/dL, ALT/AST ≤5 x ULN
No stated restriction
doxorubicin-sorafenib vs. doxorubicin 19.1% vs. 10.2% any-grade. N = 2 G3/ 4 GI bleeding in the doxorubicin-sorafenib group.
AP study (21)
Sorafenib vs. Placebo
CP A albumin ≥28 g/L; total bilirubin ≥51·3 μmol/L; ALT ≤5 x ULN
Bleeding in previous 30 days excluded.
4(2.7%) vs. 3(4%) upper GI hemorrhage– all G1 or G2. No G3/4 hemorrhage
Abou-alfa et al. (25)
CP A/B 72%/28%; minimal other details given;
No comment on ex criteria re: varices or bleeding.
1 death from intracranial hemorrhage No other bleeding events reported
PFS or TTP used as surrogate for period of follow-up evaluation unless stated.
Lipos dox/ gemcitabine
Duration disease control
Etoposide, cisplatin, doxorubicin, FU/LV
Gemcitabine, oxaliplatin, cetuximab
There was marked variability across all the studies with regard to the stated laboratory eligibility criteria for entry onto the study. As illustrated in Table 2, the required platelet count ranged from 40 to 150 and international normalization ratio (INR) from 1 to 2.3. The two largest studies, the SHARP and AP studies, allowed for a platelet count of greater than or equal to 60,000. Six of the studies restricted entry to Childs-Pugh A patients only. No study was identified which mandated endoscopy to exclude patients with varices. However, after the occurrence of gastrointestinal hemorrhage in the course of two studies3, 4 (in one case, a fatal variceal hemorrhage), the protocol was amended to require screening endoscopy prior to inclusion in the study in patients with any evidence of portal hypertension. Patients found to have esophageal varices on screening examination were eligible for the study following adequate treatment with banding or sclerotherapy and repeat endoscopy showing the varices to be obliterated, minimal, or grade 1.
Effect of Sorafenib on Odds of Bleeding Event as Compared to Control in HCC Randomized Studies.
There were four HCC randomized studies involving sorafenib. The forest plot in Fig. 1A visually depicts the pooled overall estimate of the effect of sorafenib on all bleeding events within HCC randomized studies. The number of events for the sorafenib arm was 48/722 (6.65%) and was 22/653 (3.37%) for the control arm, giving an OR of 1.77 (95% CI 1.04, 3.0) for both the fixed and random effects models. This result provides evidence of a significant (P = 0.04) increase in the odds of bleeding events (all grades) with sorafenib compared to control. Figure 1B visually depicts the pooled overall estimate of the effect of sorafenib on grades 3-5 bleeding events within HCC randomized studies. The number of events for the sorafenib arm was 33/722 (4.57%) and 13/653 (1.99%) for the control arm, giving an OR (95% CI) of 1.76 (0.91, 3.41) for both the fixed and random effects models. This result provides evidence that sorafenib did not significantly increase (P = 0.11) the odds of bleeding events (grades 3-5) as compared to a non-antiangiogenic control.
Odds Ratios for Bleeding Event Between Antiangiogenic and Non-Antiangiogenic Single-Arm Studies in HCC.
To examine the risk of bleeding event in antiangiogenic therapy compared to non-antiangiogenic therapy among single-arm studies in HCC, 19 studies incorporating antiangiogenic therapy and 21 with non-antiangiogenic therapy (Tables 2, 3) were analyzed. Figure 2 shows that, among single-arm HCC studies, the OR for any bleeding event with antiangiogenic therapy is 4.34 (2.16, 8.73; P < 0.0001). The OR of bleeding event grades 3-5 for antiangiogenic therapy are 2.66 (95% CI 1.03, 6.82; P = 0.0425). This suggests that antiangiogenic therapy significantly increases the odds of bleeding events (both all grades and grades 3-5) as compared to non-antiangiogenic therapy in single-arm HCC studies.
Effect of Sorafenib on Odds of Bleeding Event as Compared to Control in RCC and HCC Randomized Studies.
In order to determine if the observed trend towards increased hemorrhagic risk was inherent to HCC or was a class effect, we examined the effect of sorafenib on bleeding events in RCC (Fig. 3). Among the RCC randomized studies, treatment with sorafenib significantly increased the odds of any bleeding event (OR 1.92; 95% CI 1.30, 2.85) compared to control. The test for subgroup differences showed the effect of sorafenib on any bleeding event to be similar between the HCC and RCC subgroups (P = 0.75). Similar to the HCC result, treatment with sorafenib did not significantly increase the odds of bleeding events grades 3-5 (OR 1.18; 95% CI 0.58 to 2.38) among the RCC randomized studies. The overall pooled estimate of HCC and RCC studies also indicates a nonsignificant effect of sorafenib on bleeding events grades 3-5 (OR 1.43; 95% CI 0.88, 2.32), which was similar for both HCC and RCC subgroups (P = 0.45).
Worldwide, HCC is the fifth most common malignancy, with a median survival of 6-9 months.5 In the United States the incidence of HCC continues to rise, a trend which will likely result in more clinical trials being performed in this disease.6, 7 In addition, after decades of negative studies in HCC the SHARP and AP studies provided an impetus for the investigation of “antiangiogenic” strategies in HCC in an effort to bolster the relatively small gains made with sorafenib. We have learned however from the experience in other tumor types that anti-vascular endothelial growth factor (VEGF) therapies are associated with class toxicities, including bleeding. In one meta-analysis of bevacizumab-related toxicities, hemorrhagic events accounted for almost one-quarter of the fatal adverse events seen.8 In HCC this is a particular concern because of the almost invariable presence of cirrhosis in this patient population, placing them at an elevated baseline risk of hemorrhage. The main purpose of this analysis was to determine if there was an increased risk of bleeding for a patient with HCC taking part in a study evaluating an antiangiogenic therapy. For comparison we also performed an analysis of studies evaluating sorafenib in RCC. Both analyses show that whereas the use of sorafenib was associated with an increased risk of bleeding in HCC, this was primarily for lower-grade events and similar in magnitude to the risk encountered in RCC. Although the nature of these low-grade events for the most part were not characterized, they were defined as grade 1 or 2, meaning that by Common Terminology Criteria for Adverse Events (CTCAE) definition they did not require transfusion or endoscopic intervention. We also set out to describe the heterogeneity with regard to the eligibility criteria employed across HCC trials for study entry, as these are reflective of baseline risk and are nonstandardized in HCC compared to other solid tumors.
A major drawback of this analysis is the fact that only four of the HCC studies contained a control arm. When we confined our analysis to randomized studies—all of which involved sorafenib—there was a significant (P = 0.04) increase in the odds of bleeding events of (OR 1.73; 95% CI 1.02, 2.94) associated with sorafenib compared to control, although this increased risk appeared to be confined to lower-grade events. To ascertain whether this was a disease-specific effect we also analyzed the hemorrhagic risk in studies evaluating sorafenib in RCC and found the risk to be similarly increased (OR 1.92; 95% CI 1.30, 2.85), suggesting that this is not necessarily greater in HCC patients beyond the effect of the drug itself. Given that the majority of studies evaluating an antiangiogenic agent in HCC are single-arm, nonrandomized phase 2 trials it is very difficult to estimate the overall bleeding risk in these studies, especially given the heterogeneity of treatments administered. To address this, at least partially, we performed a comparative analysis with a group of single-arm phase 2 studies that did not include therapy considered antiangiogenic. We acknowledge the limitations of this approach, and certainly one cannot draw conclusions on causal effects from these uncontrolled studies evaluating heterogeneous agents. Because our outcome is one of safety, however, we felt that the phase II single-arm studies could provide valid additional data. The results were instructive in that patients with HCC taking part in these studies appeared to have a significantly increased risk of all-grade bleeding compared to non-antiangiogenic therapy. It must be emphasized, however, that sorafenib is the only approved agent for HCC. From a biological standpoint, although we do not know whether the benefit of sorafenib is predominantly related to its signal transduction inhibition versus its antiangiogenic properties, it seems likely—by association rather than direct proof—that the hemorrhagic risk is related to its anti-VEGF effect.9, 10 It is known that VEGF has an important role in the maintenance of architectural integrity within the endothelial cells of the microvasculature, inhibition of which may induce the increased risk of bleeding.
A related aim of our analysis was also to describe the variability in eligibility criteria used across studies with regard to both laboratory criteria—which are necessarily different from studies in other diseases—and the requirements with regard to allowing for, or ruling out, known or suspected varices. This is an important practical consideration as we attempt to improve the quality of studies in HCC and minimize risk. None of the studies mandated that an upper endoscopy be performed in order to screen for the presence of varices, although two studies did introduce this after the occurrence of a serious hemorrhage. The risk of variceal hemorrhage in patients with cirrhosis is difficult to quantify but has been reported to be as high as 40%.11 In the context of HCC the short- to medium-term risk is particularly important to assess given that in the SHARP study the median duration of sorafenib treatment was 5.3 months and the median overall survival for these patients was 10.7 months. The presence of even small varices is a marker of increased bleeding risk as shown by one prospective study where (12) the 2-year risk of bleeding was found to be significantly higher in patients with small varices at enrollment compared to those who did not have any varices (12% versus 2%). Several factors have been employed to predict the risk of variceal hemorrhage, including the size and location of varices (gastric fundus varices of higher risk13), their physical appearance, and variceal pressure as measured by endoscopic gauge.14 The North Italian Endoscopic Club (NIEC) study established a prognostic index—depending on size, presence of red wale marks, and Child class—which quantified 1-year bleeding risk, a relevant timepoint for a patient with a diagnosis of advanced HCC.15 According to that study there are “high risk” small varices (those that occur in Child C patients or have red wale marks) that may have the same risk of bleeding than a Child A patient with large varices (and prophylaxis is recommended in these patients). Limiting eligibility in HCC studies to Childs A patients—as recommended by the American Association for the Study of Liver Diseases (AASLD)16—would mitigate some of this risk, but—also consistent with current AASLD guidelines17—Child A patients should undergo screening endoscopy unless they have had one in the last 2-3 years (with no varices demonstrated) or last 1-2 years (if small varices had been identified. Five of the studies we reviewed—including both the SHARP and AP studies—were confined to patients with Childs-Pugh grade A cirrhosis. Perhaps the most specific indicator of risk for variceal bleeding is the prior occurrence of a hemorrhagic event with the risk of a subsequent bleeding episode estimated to be 17%-40%,18 but also, in older analyses, as high as 70%.19 Seven of the studies in our analysis excluded patients with a history of active bleeding, although the duration of this was variable, ranging from 30 days to 1 year. Given that there are effective therapies to prevent first variceal hemorrhage and also recurrent hemorrhage, patients enrolled in these studies should be screened and treated accordingly, with eligibility being confirmed on subsequent endoscopy—as in the study by Thomas et al.4—if the varices are found to be “obliterated, minimal, or grade 1.”
Although the risk of hemorrhagic event in studies evaluating an antiangiogenic agent in HCC appears to be not significantly raised for serious (grade 3-5) events, there are no standardized across-study eligibility criteria for this “at risk” population in terms of platelet count, prothrombin time, or endoscopic requirements. The eligibility criteria for HCC studies tend to be different from other settings to allow for the hepatic dysfunction that is generally present. For example, the SHARP study required a platelet count of greater than 60,000. Future studies will need to address this issue in more detail, particularly when multiple vascular targeting agents are combined.
In summary, this analysis of both randomized and nonrandomized studies evaluating an antiangiogenic agent in HCC showed that, whereas the use of sorafenib was associated with an increased risk of bleeding in HCC, this was primarily for lower-grade events and similar in magnitude to the risk encountered in RCC.
We thank Tito Fojo for helpful comments. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government.