Reliable prediction of survival in advanced-stage hepatocellular carcinoma treated with sorafenib: Comparing 1D and 3D quantitative tumor response criteria on MRI

Abstract

Objectives

To compare 1D and 3D quantitative tumor response criteria applied to DCE-MRI in patients with advanced-stage HCC undergoing sorafenib therapy to predict overall survival (OS) early during treatment.

Methods

This retrospective analysis included 29 patients with advanced-stage HCC who received sorafenib for at least 60 days. All patients underwent baseline and follow-up DCE-MRI at 81.5±29.3 days (range 35–140 days). Response to sorafenib was assessed in 46 target lesions using 1D criteria RECIST1.1 and mRECIST. In addition, a segmentation-based 3D quantification of absolute enhancing lesion volume (vqEASL) was performed on the arterial-phase MRI, and the enhancement fraction of total tumor volume (%qEASL) was calculated. Accordingly, patients were stratified into groups of Disease Control (DC) and Disease Progression (DP). OS was evaluated using Kaplan-Meier curves with log-rank test and Cox proportional hazards regression model.

Results

The Kaplan-Meier analysis revealed that stratification of patients in DC vs. DP according to mRECIST (p=0.0371) and vqEASL (p=0.0118) successfully captured response and stratified OS, while stratification according to RECIST and %qEASL did not correlate with OS (p=0.6273 and p=0.7474, respectively). Multivariable cox regression identified tumor progression according to mRECIST and qEASL as independent risk factors of decreased OS (p=0.039 and p=0.006, respectively).

Conclusions

The study identified enhancement-based vqEASL and mRECIST as reliable predictors of patient survival early after initiation of treatment with sorafenib. This data provides evidence for potential advantages 3D quantitative, enhancement-based tumor response analysis over conventional techniques regarding early identification of treatment success or failure.

Introduction

Hepatocellular carcinoma (HCC) is the 3^rd most common cause of cancer-related death and is often diagnosed at intermediate to advanced stages with a poor overall prognosis [1]. For patients in stage C of the Barcelona Liver Cancer Staging System (BCLC) systemic treatment is the recommended option [2] based on data from the seminal “Sorafenib HCC Assessment Randomized Protocol” (SHARP) and Asian-Pacific trials [3,4]. Sorafenib was the first systemic agent to be approved for advanced HCC in 2007 and remained the only approved drug until 2017, when regorafenib was established as FDA-approved second-line treatment for patients that failed or progressed on sorafenib [5], followed by lenvatinib as alternative TKI-based first-line therapy [6]. Additional drugs followed in short sequence: immune checkpoint inhibitors nivolumab, pembrolizumab, ramucirumab and TKI cabozantinib as single-agent therapy [7–10], as well as multi-drug combinations like nivolumab and ipilimumab [11] are now approved for HCC. However, sorafenib still remains a common first line treatment in most protocols.

While many systemically applicable agents offer promise of achieving survival benefits, no early imaging biomarkers exist that accurately assess efficacy and tumor response in early stages of therapy. Given the heterogeneity of HCC and variable response rates to systemic therapy, tools to reliably identify progression would provide rationale for initiation of second-line therapy. Furthermore, current clinical trials on HCC treatment implement the concept of sequential use of different systemic agents or combination of systemic therapy with locoregional therapy [12]). Reliable and early predictors of response to TKI are needed in those investigations and might help to expand availability of systemic treatment to HCC patients in earlier stages. Currently accepted tumor response assessment techniques in HCC include one-dimensional (1D) response evaluation criteria in solid tumors (RECIST 1.1) and modified RECIST (mRECIST), an enhancement-based variant of the previously established standard [13,14]. Several studies indicate that RECIST is unable to accurately assess response to molecularly targeted agents such as sorafenib since it merely reflects on changes in total tumor size neglecting therapy-induced intra-tumoral changes such as necrosis or enhancement alterations [15,16]. The enhancement-based mRECIST was introduced to address those limitations and showed to be more effective in predicting overall survival (OS) after systemic therapy [17, 18]. While mRECIST represents an incremental improvement, it is widely understood that assessing response on a single axial plane is highly reader-dependent and does not reflect changes of the entire, frequently complex tumor volume and shape. In order to overcome those inherent limitations, the three-dimensional (3D) enhancement-based quantitative European Association for the Study of the Liver (qEASL) technique has been developed and validated as capable of assessing tumor response in HCC patients undergoing loco-regional therapy with greater accuracy than any currently available 1D criteria [17,19,20]. Previous studies have proven findings of qEASL to correlate well with histopathological results [21] and demonstrated its feasibility in a workflow-efficient time frame [22].

This study aims to compare 1D and 3D quantitative tumor response criteria on dynamic contrast-enhanced MRI (DCE-MRI) in patients with advanced-stage HCC undergoing systemic treatment with sorafenib, and to evaluate the ability of those techniques to predict OS as the primary outcome endpoint.

2. Methods

This is a retrospective single institution analysis which was approved by the institutional review board and is compliant with the Health Insurance Portability and Accountability Act. Informed consent was waived. Study design was in agreement with the Standards for Reporting of Diagnostic Accuracy guidelines.

2.1. Study Population

A systematic search through electronic health records helped us to identify 238 HCC patients who received systemic treatment with sorafenib between October 2011 and September 2019. A total of 209 patients had to be excluded because of insufficient imaging (n=93), treatment with sorafenib for less than 60 days (n=60), additional loco regional treatment of the target lesions prior or along with sorafenib treatment (n=30) or absence of a verifiable target lesion (n=26). Twenty-nine patients with intermediate (BCLC B) or advanced (BCLC C) HCC who received systemic treatment with sorafenib (Bayer Healthcare) for at least 60 days within our institution were included in the study. In 19 patients HCC was biopsy proven, diagnosis in the remaining 10 patients was established in concordance with LI-RADS criteria [23]. All included patients had inoperable HCC, did not receive prior systemic treatment and underwent DCE-MRI at baseline (BL) before and on at least one follow-up (FU) time point after initiation of therapy with sorafenib. The first follow-up imaging was acquired at 81.5±29.3 days (median±SD, range 35–140 days). Up to two dominant lesions were analyzed for each patient.

2.2. Sorafenib treatment

In patients tolerating sorafenib, the standard recommended daily dose of 800mg (400mg twice daily) was continuously administered orally. In cases of partial intolerance, the treatment plan was subject to dose reductions to 400mg or 200mg daily or interruptions due to side effects such as hand-foot-syndrome, diarrhea or other forms of toxicity. Patients that interrupted systemic treatment for longer than 48 hours between initiation of sorafenib and first follow-up time point was not included in our study.

2.3. MRI Imaging protocol

All patients underwent DCE-MRI at baseline and follow-up. Images were obtained using a clinical 1.5-T MR imager (Magnetom Avanto; Siemens) and a phased array torso coil. The imaging protocol included axial T2 weighted fast spin-echo images, axial T1 weighted dual fast gradient echo images and axial breath-hold unenhanced and contrast-enhanced T1 weighted images (0,1 mmol per kilogram of body weight intravenous gadodiamide (Dotarem; Guerbet). All contrast-enhanced images were acquired in arterial, portal venous and delayed phases (20, 70 and 180 seconds after intravenous contrast administration, respectively).

2.4. Response Assessment

Response to sorafenib was retrospectively assessed in 46 lesions of 29 patients according to 1D and 3D criteria. Up to two target lesions were analyzed for each patient. In cases with multiple lesions the tumors with the largest axial diameters at baseline were chosen as target lesions. 1D analysis was performed measuring the longest diameter (RECIST 1.1) or longest enhancing diameter (mRECIST) on arterial phase MR images at baseline and respective follow-up time points. 3D analysis using qEASL was conducted as previously described (22). Briefly, first a 3D semiautomatic tumor segmentation was performed on arterial phase DCE-MRI images using a 3D tumor segmentation software (IntelliSpace Portal V8, Philips Healthcare). Previous studies evaluated inter-reader liability and pathologic validation for the 3D tumor segmentation (19, 21). The algorithm used pre-contrast phase images to be registered to and subtracted from the corresponding arterial phase image in order to eliminate any pseudo-enhancement that was present on pre-contrast phase. Subsequently, a cubic (1 cm³) region of interest (ROI) was chosen on non-tumorous liver tissue, if possible, ipsilateral to the target lesion avoiding visible structures like blood vessels (21). Viable enhancing tumor was defined as voxels within segmented areas that showed greater voxel intensity than the mean plus two standard deviations of the voxel intensity in the ROI. Clinical feasibility and time efficiency in assessment of qEASL were previously demonstrated (22).

Applying those analyses to both BL and FU images, two different qEASL measures were calculated. Volume-based qEASL (vqEASL) conveys the percentage change in absolute enhancing tumor volume (eTV) in cubic centimeters between baseline (BL) and follow-up (FU) ($\mathit{\text{vqEASL}}=\frac{eTV(FU)-eTV(BL)}{eTV(BL)}*100[\text{\%}]$). Percent-based qEASL (%qEASL) compares the fraction of eTV in total tumor volume (TTV) between BL and FU and also expresses percent change ($\text{\%}\mathit{\text{qEASL}}\mid =\frac{\frac{eTV(FU)}{TTV(FU)}-\frac{eTV(BL)}{TTV(BL)}}{\frac{eTV(BL)}{TTV(BL)}}*100[\text{\%}]$) (20). All response criteria were assessed at BL and all available FU timepoints. An exemplary assessment of all response criteria is portrayed in Figure 1.

Fig.1.

Images of a 73-year-old HCC patient with an OS of 33 months after start of Sorafenib therapy. At the first FU (middle column) the 1D criteria RECIST and mRECIST displayed in the first and second row categorized response as Disease Progression (+32% and +35%, respectively), while 3D criteria assessed response as Disease Control (%qEASL −11%, vqEASL −28%). At the second FU time point its demonstrated that while vqEASL and %qEASL further decreased, this is still not equally captured by RECIST and mRECIST.

2.5. Stratification into Disease Control and Disease Progression

At the time of the follow-up scan, patients were stratified into Disease Control (DC) including all patients with complete response (CR), partial response (PR) and stable disease (SD) and Disease Progression (DP) according to each of the four response assessment tools. For 1D criteria an increase higher than 20% was defined as DP. An increase of 20% in a 1D plane is equivalent to an increase of 73% in an 3D scenario according to the formula for spherical volume = 4/3πr³. Analogously, for 3D criteria such as vqEASL and %qEASL an increase of higher than 73% was defined as disease progression. Those thresholds were previously validated to measure response after locoregional treatments [24, 25]. Due to the less invasive and more gradual anti-angiogenetic effect of systemic therapy in comparison to intra-arterial therapies, changes in radiological appearance and enhancement uptake might be more discrete. Taking this into consideration a second analysis with different cut-off values was conducted. For RECIST and mRECIST criteria any increase in (enhancing) diameter was defined as DP, for 3D criteria any increase of eTV (vqEASL) or fraction of eTV in TTV (%qEASL) between BL and FU was defined as PD.

2.6. Statistical analysis

Continuous variables are expressed as means and standard deviations, categorical variables as percentages. Median overall survival was determined as time period between beginning of systemic treatment with sorafenib and death of any kind. Patients lost to follow-up or still alive were censored at the last follow-up time point or the end of the observation period. Survival analysis was performed using Kaplan-Meier method and log-rank test. To assess independent predictors of response to Sorafenib a two-step approach was used. In a first step, a univariate Cox regression model was used to evaluate the association of OS with clinical baseline factors such as gender, age, ECOG (Eastern Cooperative Oncology Group) performance status, number of lesions, size and volume of largest target lesion, number of pre-treatments, presence of portal vein thrombus (PVT), and time between diagnosis and start of sorafenib treatment. In a second step, variables that correlated with response to Sorafenib (p<0.1) were included in a multivariable cox regression model to estimate adjusted hazard ratios (HR). Statistical analysis was performed using Prism (v7.0, GraphPad) and SPSS (Version 26, IBM).

3. Results

3.1. Patient Characteristics

Baseline Characteristics are summarized in Table 1. Reflecting on a higher prevalence of HCC in male patients, 82.8% of participants were male, mean age was 63±8.26 (mean ± SD). 89% of patients underwent loco-regional therapy of non-target lesions, the mean number of previous treatments was 2.59±1.79. Mean OS was 13.4±9.46 months (range 5–43 months). Mean lesion size at BL was 2.96±2.75 cm. Mean time of treatment with sorafenib was 6±5.6 months. The median follow-up period was 9.6 months. At the end of the observation period (09/2019) 17 patients (58%) were deceased.

Table 1.

Baseline Patient Characteristics

Parameter	N (%)
Demographics
Number of Patients	29 (100%)
Age (years), mean and standard deviation	63 ± 8.26
Male/ Female	24 (82.8) / 5 (17.2)
Ethnicity
Caucasian	22 (75.9)
African-American	1 (3.4)
Hispanic	6 (20.7)
Disease Characteristics
Cirrhosis	27 (93.1)
Etiology of Cirrhosis
Hepatitis C	12 (44.4)
Alcohol consumption	5 (18.5)
Nonalcoholic steatohepatitis (NASH)	5 (18.5)
Hepatitis C and Alcohol Consumption	4 (14.8)
NASH and Alcohol consumption	1 (3.7)
ECOG Performance Status
0	4 (13.8)
1	23 (79.3)
2	2 (6.9)
Child Pugh class
A	18 (62.1)
B	11 (37.9)
BCLC
B	1 (3.4)
C	28 (96.6)
Main portal vein thrombosis	10 (34.5)
Previous treatment
Pre-treated/ not pre-treated	26 (89.7) / 3 (10.3)
Number of previous treatment, mean and standard deviation	2.59 ± 1.79
Resection	3 (11.5)
Loco-regional treatment	14 (53.9)
Y90	1 (3.9)
Resection and loco-regional treatment	3 (11.5)
Y90 and loco-regional treatment	5 (19.2)

ECOG, Eastern Cooperative Oncology Group; BCLC, Barcelona Clinic Liver Cancer; Y90, Radio embolization with Yttrium-90

3.2. Response analysis and survival

The survival analysis using previously validated cut-off thresholds is shown in Figure 2. The ability of the different response criteria to stratify patients into DC and DP according to observed OS was evaluated. RECIST and %qEASL failed to stratify patients into DC and DP (DC n=24, DP n=22, p=0.6273 and DC n=44 DP=2, p=0.7474, respectively). A stratification according to mRECIST (DC n=31, DP n=15) was significantly associated with OS (p=0.0371) and also vqEASL was able to accurately differentiate between DC and DP (DC n=29, DP n=17, p=0.0118).

Fig.2.

Survival analysis based on target lesion response measured according to (A) RECIST, (B) mRECIST, (C) %qEASL and (D) vqEASL using previously validated cut-off thresholds at timepoint of first FU after start of systemic therapy with sorafenib

Two separate multivariable cox regression models were created for mRECIST and vqEASL since both are enhancement-based markers. After adjusting for gender and presence of PVT, DC according to mRECIST (HR=0.325, p=0.039. 95%CI 0.112–0.946) or vqEASL (HR=0.183, p=0.006, 95%CI 0.055–0.613) were proven independent prognostic markers for increased OS.

In a second survival analysis patients were stratified according to absolute changes in the respective response criteria. Fig. 3A,C,E,G presents waterfall plots displaying percentual changes of each criterium. In a second survival analysis stratification was based upon absolute decrease in each respective criterium vs. absolute increase. Between BL and the first FU, RECIST decreased in n=15 and increased in n=31 lesions, %qEASL was decreased in n=24 and increased in n=22 lesions. Both RECIST (Fig.3B) and %qEASL (Fig.3F) failed to adequately classify patients into DC and DP (p=0.2009 and p=0.2775 respectively). Comparing absolute decrease of mRECIST (n=16) versus increase (n=30) and absolute decrease in vqEASL (n=14) versus absolute increase (n=32), a significant association with OS was noted (Fig.3D, H; p=0.0205 and p=0.0098 respectively). These results indicate that only response assessment according to mRECIST and vqEASL is associated with survival outcomes.

Fig.3.

Waterfall Plot demonstrating change in (A) RECIST (C) mRECIST, (E) %qEASL and (G) vqEASL for each analyzed lesion (n=46) and survival analysis stratifying according to absolute changes on (B) RECIST (D) mRECIST (F)%qEASL and (H) vqEASL

3.3. Long-term development of eTV

Fig.4 displays a spider plot showing growth and shrinkage of enhancing tumor tissue. Using vqEASL analysis eTV was assessed in all target lesions in all available FU images. The percentual change from BL (100%) over the course of the FU time points is displayed for each target lesion.

Fig.4.

Spider plot demonstrating changes in eTV from baseline over all available FU time points for each lesion (n=46) and stratification into DC (eTV increase <73%, n=29) and DP (eTV increase ≥73%, n=17) according to vqEASL at 1^st FU

Discussion

This retrospective analysis suggests that enhancement-based imaging biomarkers mRECIST and vqEASL are reliable predictors of patient survival after sorafenib therapy in patients with advanced-stage HCC. At the same time, we found that RECIST1.1 and %qEASL were unable to effectively stratify between DC and DP during treatment with sorafenib.

Sorafenib is a small-molecular multi-kinase inhibitor which targets pathways involved in both cell proliferation and tumor angiogenesis [3, 4, 26], ultimately acting to inhibit tumor cell growth and disrupt its microvascular blood supply. The therapy cost is exceedingly high and only a small proportion of patients show sustainable response justifying the currently unmet need for reliable markers to distinguish patients benefiting from and those failing on sorafenib. Numerous clinical, biological and radiological markers have been tested for association with prognosis. While promising biological markers like occurrence of adverse events, macrovascular invasion and plasma angiopoetin-2 levels have been identified, strong radiological predictors are still missing [27].

Consistent with previous data [15,16,25], we corroborate that RECIST1.1 is a poor marker of early treatment response which does not reflect on complex and gradually occurring intratumoral changes in enhancement patterns therapy induced by the rather cytostatic than cytotoxic effects of Sorafenib. Therefore, we conclude that RECIST1.1 should not be clinically used for monitoring of HCC patients receiving sorafenib.

1D assessment of tumor enhancement on a single axial plane is reader-dependent, may not be representative of the entire tumor volume and may disregard asymmetrical, multi-compartmental tumor growth or response [19, 20, 22, 25, 28]. However, in agreement with Lencioni et al. [18] and Kudo et al [17], our study showed that tumor response assessment using mRECIST is an independent prognostic factor for OS. This enhancement-based approach is able to detect devascularization induced by sorafenib, but several limitations remain. mRECIST is obtained on a single plane which may be misleading as it implies that changes in tumor viability are homogeneous throughout the entire tumor volume. On the contrary, treatment with sorafenib often results in irregular and heterogeneous patterns of enhancement at an early stage of therapy followed by complex patterns of intra-tumoral necrosis without measurable tumor shrinkage [29,30].

Previous studies investigated HCC tumor response to loco-regional therapies and demonstrated that computer-assisted semiautomated segmentation-based 3D volumetric response assessment may be advantageous in overcoming the aforementioned limitations of 1D mRECIST. qEASL has also been shown to have higher inter-reader reproducibility and lower variability with respect to classification of patients as responders vs. non-responders [22,28,31]. When applied to systemic therapies, those findings are reinforced by the results of our study which confirmed that progression identified using vqEASL is an independent prognostic marker for OS with a higher HR than progression in mRECIST and therefore shows better correlation with survival outcome. While technically more complex, this 3D technique is biologically more accurate and reproducible and provides a more wholistic evaluation of the entire tumor at any time point after initiation of therapy with sorafenib [21].

In our study, we also found that stratification of patients as responders (CR and PR) vs. non-responders (SD and PD) did not result in any significant separation in outcome predictions and may not be advantageous to faithfully compare different tumor response instruments regarding their predictive potential. On the contrary, when comparing DC (CR, PR and SD) vs. DP (including PD alone), nearly all applied criteria demonstrated some degree of separation in predicted OS with best results in enhancement-based markers. This might be explained by a very low number of complete or partial response observed in patients with such advanced stages receiving sorafenib and that a separation between patients not showing progression (DC) vs. those with progression (DP) might be more clinically suitable.

Numerous studies on radiological response assessment in Sorafenib have been published, but only few have examined radiological changed in tumor morphology such as loss of vascularity reflected in less enhancement or appearance of necrosis. Response identified by Choi criteria are based on reduction of longest lesion diameter and volumetric tumor attenuation and have been proven to be associated with increased survival in HCC patients treated with sorafenib [32,33]. However, Choi criteria in sorafenib response assessment have only been tested on CT so far and inter-reader agreement has been found to be moderate [33]. Furthermore, increase in apparent diffusion coefficient values obtained using diffusion-weighted MRI might be associated with favorable response to sorafenib [34, 35], but reliable data are still lacking.

There were several limitations to our study. First, our data was retrospective in nature and used a relatively small patient cohort, of which a majority had received prior treatment in non-target lesions. Moreover, the fairly wide range of time period between BL and FU imaging should be recognized. While reflective of true clinical reality with irregularities in scheduling of FU imaging sessions after initiation of sorafenib treatment, statistical significance of 3D response assessment techniques in our study demonstrates its robustness, making it more likely to be applicable to clinical practice as a decision support tool.

In summary, this comparison of MRI-based 1D and 3D quantitative response assessment techniques in patients with advanced-stage HCC indicates that vqEASL has the highest predictive value while mRECIST continues to be a sufficiently accurate predictor of overall survival early after initiation of therapy with a TKI. This data additionally provides additional evidence for advantages of enhancement-based 3D quantitative tumor response analysis over conventional 1D techniques regarding early detection of non-response to sorafenib. It can thus be suggested to use vqEASL as a clinical decision support tool when considering changes from sorafenib to other second-line systemic therapies in treatment of advanced-stage liver cancer. Furthermore, 3D response assessment tools as vqEASL can be beneficial as early response markers in current clinical research trials investigating a combination of different systemic agents or systemic agents and locoregional therapy, which overall might lead to an expanded availability of systemic therapies to HCC patients in earlier stages.

Key points :

• Tumor response criteria on MRI can be used to predict survival benefit of sorafenib therapy in patients with advanced HCC.

• Stratification into DC and DP using mRECIST and vqEASL significantly correlates with OS (p=0.0371 and p=0.0118, respectively) early after initiation of sorafenib, while stratification according to RECIST and %qEASL did not correlate with OS (p=0.6273 and p=0.7474, respectively).

• mRECIST (HR=0.325, p=0.039. 95%CI 0.112–0.946) and qEASL (HR=0.183, p=0.006, 95%CI 0.055–0.613) are independent prognostic factors of survival in HCC patients undergoing sorafenib therapy.

Abbreviations

%qEASL

percent-based qEASL

1D

one-dimensional

3D

three-dimensional

95%CI

95% Confidence Interval

BCLC

Barcelona Liver Cancer Staging System

BL

Baseline

CR

Complete Response

DC

Disease Control

DCE-MRI

Dynamic contrast-enhanced magnetic resonance imaging

DP

Disease Progression

ECOG

Eastern Cooperative Oncology Group

eTV

enhancing tumor volume

FDA

U.S. Food and Drug Administration

FU

Follow-Up

HCC

Hepatocellular carcinoma

HR

Hazard Ratio

LI-RADS

Liver Imaging Reporting and Data System

OS

Overall Survival

PR

Partial Response

PVT

portal vein thrombus

qEASL

quantitative European Association for the Study of the Liver

SD

Stable Disease

TKI

tyrosine kinase inhibitors

TTV

total tumor volume

VEGFR2

Vascular Endothelial Growth Factor Receptor 2

vqEASL

volume-based qEASL