Targeted Molecular Profiling of Circulating Cell-Free DNA in Patients With Advanced Hepatocellular Carcinoma

NGS of cfDNA in HCC represents an alternative to tissue-based profiling, given the high tumor-plasma NGS concordance.

Abstract

PURPOSE

Next-generation sequencing (NGS) of tumor-derived, circulating cell-free DNA (cfDNA) may aid in diagnosis, prognostication, and treatment of patients with hepatocellular carcinoma (HCC). The operating characteristics of cfDNA mutational profiling must be determined before routine clinical implementation.

METHODS

This was a single-center, retrospective study with the primary objective of defining genomic alterations in circulating cfDNA along with plasma-tissue genotype agreement between NGS of matched tumor samples in patients with advanced HCC. cfDNA was analyzed using a clinically validated 129-gene NGS assay; matched tissue-based NGS was analyzed with a US Food and Drug Administration–authorized NGS tumor assay.

RESULTS

Fifty-three plasma samples from 51 patients with histologically confirmed HCC underwent NGS-based cfDNA analysis. Genomic alterations were detected in 92.2% of patients, with the most commonly mutated genes including TERT promoter (57%), TP53 (47%), CTNNB1 (37%), ARID1A (18%), and TSC2 (14%). In total, 37 (73%) patients underwent paired tumor NGS, and concordance was high for mutations observed in patient-matched plasma samples: TERT (83%), TP53 (94%), CTNNB1 (92%), ARID1A (100%), and TSC2 (71%). In 10 (27%) of 37 tumor-plasma samples, alterations were detected by cfDNA analysis that were not detected in the patient-matched tumors. Potentially actionable mutations were identified in 37% of all cases including oncogenic/likely oncogenic alterations in TSC1/2 (18%), BRCA1/2 (8%), and PIK3CA (8%). Higher average variant allele fraction was associated with elevated alpha-fetoprotein, increased tumor volume, and no previous systemic therapy, but did not correlate with overall survival in treatment-naïve patients.

CONCLUSION

Tumor mutation profiling of cfDNA in HCC represents an alternative to tissue-based genomic profiling, given the high degree of tumor-plasma NGS concordance; however, genotyping of both blood and tumor may be required to detect all clinically actionable genomic alterations.

INTRODUCTION

Advanced hepatocellular carcinoma (HCC) is a disease with high morbidity and mortality, and—even with the advent of effective immunotherapy—the 5-year overall survival (OS) for advanced disease remains poor.^1-3 Genomic profiling has limited impact on treatment decisions in patients with advanced HCC, which contrasts with other solid tumors such as biliary tract cancers, where tumor molecular profiling is an established component of standard care as a guide to eligibility for a variety of US Food and Drug Administration (FDA)–approved precision medicines.^4-6 As HCC may be diagnosed based on imaging criteria alone, pretreatment tissue is often unavailable, and when tissue is acquired, may be of limited quality and/or quantity because of technical issues such as concerns about the safety of tumor biopsy in patients with co-occurring cirrhosis.^7-9 As such, predictive and prognostic genomic-based biomarkers remain largely underexplored in the disease.¹⁰

CONTEXT

• Key Objective

• Reports of molecular profiling of cell-free DNA (cfDNA) in hepatocellular carcinoma (HCC) to date have been limited by heterogenous patient populations and variability of confirmatory tissue diagnosis making the results challenging to interpret. We evaluated a panel-based plasma cfDNA assay in a histologically confirmed cohort of patients with advanced HCC to define the genomic landscape in cfDNA and determine plasma-tissue concordance.

• Knowledge Generated

• In this analysis, there was a high rate of detection of genetic alterations in cfDNA with a high level of concordance in matched tissue samples. Driver alteration detection frequency in plasma was also not different to historic tissue sequencing cohorts.

• Relevance

• The high detection rate of genetic alterations in cfDNA with overall high concordance supports the use of plasma-based molecular profiling where tissue acquisition is not feasible. Blood and tumor sequencing may be required to detect all clinically relevant and potentially actionable alterations.

A potential less-invasive alternative to tumor genotyping in HCC may come from the ability to detect and analyze using next-generation sequencing (NGS)–based methods, tumor-derived cell-free DNA (cfDNA).^11,12 Cancers, including HCC, have been shown to shed cfDNA into the blood, which can be isolated and analyzed using a variety of methods, including targeted digital polymerase chain reaction for individual hotspot mutations or broader mutational profiling using NGS-based methods. In addition to being minimally invasive, genotyping of tumor-derived cfDNA may have the additional advantage of capturing tumor genomic heterogeneity both spatially and temporally.^13,14 Indeed, spatial heterogeneity of somatic alterations may lead to an incomplete assessment of the mutational profile of patients with advanced cancer if only a single site of disease is biopsied.^12,13,15

To date, several studies have confirmed that detection and analysis of cfDNA is feasible in patients with HCC; however, studies of targeted NGS of patients with advanced disease are limited by either lack of clinical annotations, failure to confirm HCC with a diagnostic biopsy, and/or lack of paired tumor samples to assess concordance of tumor and plasma mutational profiles.^14,16-19 To address the limitations of previous studies, we performed a retrospective, biospecimen procurement study in patients with histologic-confirmed HCC to determine the genomic landscape of patients with advanced HCC as assessed via molecular profiling of cfDNA and the concordance of plasma and tumor analyses using paired samples collected from individual patients with HCC.

METHODS

Study Design

This was a single-center, retrospective study with the primary objective of describing the landscape of genomic alterations via NGS of cfDNA extracted from patients with histologically confirmed advanced HCC. Secondary objectives included assessing genotype agreement between sequencing from cfDNA extracted from blood and DNA extracted from tumor, the frequency of clinical actionability of genomic alterations identified by cfDNA analysis, and correlation of cfDNA concentration/variant allele fraction (VAF) with clinicopathologic factors and patient outcomes.

From February 2019 to August 2021, patients treated at Memorial Sloan Kettering Cancer Center (MSKCC) underwent cfDNA extraction followed by cfDNA sequencing using the 129-gene NGS-based MSK-Analysis of Circulating cfDNA to Examine Somatic Status (MSK-ACCESS) assay.²⁰

The electronic medical record was reviewed to obtain demographic, clinicopathologic, and treatment information. All samples were collected after obtaining written informed consent on an institutional review board (IRB)–approved research protocol (ClinicalTrials.gov identifier: NCT01775072).

Tumor Radiographic Volume

Radiographic images via magnetic resonance imaging or computer tomography obtained at a median of 13 days from cfDNA sample acquisition (0-46) were retrospectively reviewed by board-certified radiologists (T.H.K. and R.K.G.D.) who were blinded to the study data. A maximum number of five of the largest HCC nodules in each organ (ie, liver and lung) were measured when multiple lesions were present. Tumor volume of each HCC nodule was calculated according to the following mathematical equations. Tumor volume (cm³) = 4/3 × 3.14 × (maximum radius of the tumor nodule in cm).³ Liver tumor volume was calculated as the sum of the volumes of all tumor nodules within the liver. Total tumor volume (TTV) was the sum of the tumor volumes of every tumor nodule, including both hepatic and extrahepatic nodules: TTV (cm³) = tumor volume of (tumor nodule 1 + tumor nodule 2 +…tumor nodule N).²¹

Molecular Profiling

All patients underwent at least one peripheral blood collection using cfDNA blood collection tubes (Streck, La Vista, NE). Matched leukocyte DNA was used as normal control to identify germline variants and alterations consistent with clonal hematopoiesis. Extracted cfDNA was subjected to NGS testing using the MSK-ACCESS assay, a state-approved hybridization and deep sequencing assay with approximately 200,00× raw coverage and error suppression through the use of dual bar codes and collapsing of read pairs.²⁰ MSK-ACCESS detects genomic alterations in 129 cancer-associated genes (Appendix Table A1). When available, tumor tissue was analyzed using MSK-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT),⁵ an FDA-authorized hybridization capture-based targeted NGS array.²² OncoKB, a precision oncology knowledge base of clinically actionable variants, was used to identify potentially actionable genomic alterations.²³ Genes were grouped into pathways using curated templates from The Cancer Genome Atlas Program (TCGA).²⁴

Average VAF was calculated as the sum of the number of reads containing detected variants in MSK-ACCESS divided by the sum of the overall coverage at those loci for which variants were identified. For samples with previous MSK-IMPACT results, this figure was averaged with a corresponding calculation for genes for which mutations were both detected in MSK-IMPACT and covered by the MSK-ACCESS design to generate a final VAF.

All data are available for visualization and analysis via the cBioPortal for Cancer Genomics.²⁵

Statistical Analysis

Categorical data were summarized as frequencies with percentages and continuous data as medians and ranges. To examine the concordance of mutation calling between tissue and plasma (where available), we measured the percent agreement. Agreement was defined as both platforms calling the alteration or not calling the alteration. Cohen's kappa was used to measure the agreement between the two platforms. The following classification has been suggested to interpret the strength of the agreement on the basis of the Cohen's kappa value: <0 = poor, 0.01-0.20 = slight, 0.21-0.40 = fair, 0.41-0.60 = moderate, 0.61-0.80 = substantial, and 0.81-1.00 = almost perfect.^26,27

A linear regression model was used to examine a set of baseline disease characteristics (disease etiology, stage of disease, alpha-fetoprotein [AFP] ≥400 ng/mL, TTV, total liver tumor volume, extrahepatic disease, and presence of lung metastases) with cfDNA average VAF. VAF, TTV, and total liver tumor volume were log-transformed because of skewedness. Variables that were significantly associated with cfDNA VAF quantity from univariate analysis at P < .05 level were entered into the multivariable linear regression model.

A subset of patients were systemic treatment–naïve at the time of cfDNA collection (n = 24). For this cohort, OS was calculated from the date of cfDNA sample collection to the date of death or last evaluation. OS was calculated using the Kaplan-Meier methods, and a Cox proportional hazards model was used to univariately correlate ctDNA VAF (low v high, dichotomized by the median value) with OS. P values of ≤ .05 were considered significant. Statistical analysis was completed using Rv3.4.3 (The R Project for Statistical Computing, R Core Team, Vienna, Austria).²⁸

Statement of Ethics

All samples were collected after obtaining written informed consent on an IRB-approved research protocol (MSKCC; ClinicalTrials.gov identifier: NCT01775072). This study was approved by the MSKCC Institutional Review Board and was conducted in accordance with the US Common Rule.

RESULTS

Patient Demographics

Fifty-three samples from 51 patients with advanced HCC were included in the analysis. Patient demographics at the time of cfDNA collection are listed in Table 1. The median age was 69 years (range, 42-87) and 39 (76%) patients were male. Twenty-four (47%) patients had a history of hepatitis B or C, whereas 27 (53%) had a nonviral etiology (11 [22%] nonalcoholic fatty liver disease, 9 [18%] alcoholic liver disease, and 7 [13%] no identifiable risk factor).

TABLE 1.

Patient Demographics

All patients had inoperable disease or were ineligible for liver transplant at the time of plasma collection for cfDNA sequencing (American Joint Committee on Cancer stage II: 4, 8%, stage III: 9, 18%; stage IV: 38, 74%). All patients had an Eastern Cooperative Oncology Group performance status of ≤1 and Child-Pugh score for cirrhosis mortality of ≤B. Liver-limited disease was present in 17 (33%), while extrahepatic disease was present in 34 (67%) patients. Macrovascular involvement was observed in 19 (38%) patients and 16 (31%) had measurable lung metastases. Twenty-two patients had a serum AFP ≥400 ng/mL (median, 156.4; range, 1.4-1,011,863; Appendix Table A2).

Sixteen (31%) patients had previous surgery. Previous locoregional therapy had been received by 31 (61%) patients. Twenty-four (47%) patients were naïve to systemic therapy before cfDNA collection, 18 (35%) had received a tyrosine kinase inhibitor, 5 (18%) anti–PD-1/PD-L1 therapy, and 4 (8%) both targeted and immunotherapy before cfDNA collection.

Mutational Profile

Forty-eight (90.6%) of 53 plasma samples had detectable genomic alterations, with a positive sample detected in 47 of 51 (92.2%) patients. Median cfDNA yield after extraction was 39.43 ng (range, 7.93-287.68). In those samples where genomic alterations were identified, the median VAF was 0.027 (0.001-0.28). The cfDNA yield did not correlate with the VAF per sample (R² = 0.055). Including only one sample per patient, the most frequently mutated genes were TERT promoter (57%), TP53 (47%), CTNNB1 (37%), ARID1A (18%), and TSC2 (14; Fig 1). Alterations in APC, CDKN2A, and PIK3CA were detected in 8% of samples, and BRCA1/2 and NF1 alterations in 4% of samples. The most common oncogenic pathways with oncogenic or likely oncogenic alterations were the WNT-β-catenin (45%) and PIK3-AKT-TOR (25%) pathways. No samples demonstrated evidence of microsatellite instability.

FIG 1.

OncoPrint of the most frequently altered genes in 51 cfDNA samples from patients with histologically confirmed HCC. The average variant allele frequency detected per sample is highlighted in the top bars. Select clinicopathologic parameters are shown below. AFP, alpha-fetoprotein; cfDNA, cell-free DNA; HCC, hepatocellular carcinoma; VAF, variant allele fraction.

Tissue and cfDNA Mutational Profile Concordance

We compared the 47 patients from this cohort who had a positive cfDNA to a previously published, separate, cohort of patients (N = 121) who underwent real-time clinical grade genotyping with MSK-IMPACT.⁴ When comparing the frequencies of identified mutations in plasma and tissue, there was no significant difference in TERT (61% v 56%; P = .6), CTNNB1 (41% v 36%; P = .6), ARID1A (18% v 14%; P = .5), and TSC2 (16% v 9%; P = .2). There was, however, an increased prevalence of TP53 (51% v 32%; P = .024) mutations in the cfDNA cohort versus the published tissue-based HCC genomic cohort (Fig 2; Appendix Table A3).

FIG 2.

Prevalence of genomic alterations identified in the MSK cfDNA cohort compared with a previously published cohort of HCC tumors analyzed using tissue-based tumor genomic profiling.⁴ AXIN1 and BAP1 were not covered by MSK-ACCESS platform. ^aOnly one exon of JAK1 and ARID2 were covered by MSK-ACCESS. cfDNA, cell-free DNA; MSK-ACCESS, MSK-Analysis of Circulating cfDNA to Examine Somatic Status; N/A, not available.

We next compared the molecular profile between cfDNA and tumor within our 51-patient cohort to assess interassay agreement. Thirty-seven (72.5%) patients underwent tissue sequencing with a median time between tumor and cfDNA sample collection of 9.8 months (range, 0.1-70.6). Plasma cfDNA sequencing identified 92.5% of genomic alterations that had been reported clinically on tissue sequencing (Fig 3A). TERT, TP53, CTNNB1, ARID1A, TSC2, and PI3KA were detected in plasma relative to tumor in 20 of 24 (83%), 15 of 16 (94%), 12 of 13 (92%), six of six (100%), five of seven (71%), and one of two cases (50%), respectively. In 10 (27%) of 37 paired samples, additional mutations were detected by cfDNA that were not detected in the patient-matched tumor analysis (Fig 3B; Table 2). Interassay agreement between assay varied depending upon the mutation of interest: TERT (agreement 84.2, kappa = 0.67), TP53 (agreement 86.8, kappa = 0.74), CTNNB1 (agreement 92.1, kappa = 0.83), ARID1A (agreement 97.4, kappa = 0.91), TSC2 (agreement 89.2, kappa = 0.65), and PI3KA (agreement 92.1, kappa = 0.36; Appendix Table A4).

FIG 3.

(A) Fraction of genomic alterations detected (green) versus not detected (gray) in cfDNA that were previously called in tumor tissue-based genomic profiling for each individual patients who had both cfDNA (MSK-ACCESS) and tumor (MSK-IMPACT) genomic profiling. (B) Total number of alterations called in cfDNA (green) or not called (gray) that were previously identified in tissue. Orange bars indicate alterations detected in plasma but not previously identified by MSK-IMPACT analysis of tumor tissue. Of note, samples 2 and 22 were from the same patient. cfDNA, cell-free DNA; MSK-ACCESS, MSK-Analysis of Circulating cfDNA to Examine Somatic Status; MSK-IMPACT, MSK-Integrated Mutation Profiling of Actionable Cancer Targets.

TABLE 2.

Alterations Identified in cfDNA Not Identified From Matched Tissue Sequencing

Clinically Actionability

Clinically actionable mutations were identified in cfDNA in 37% of cases, including TSC1/2 (18%), BRCA1/2 (8%), and PIK3CA (8%). In 10 patient samples where genomic alterations were not detected in tissue, 4/10 (40%) had an OncoKB level of actionability of level 3b (Table 2). Two patients received everolimus in the third-line and fourth-line setting on the basis of alterations identified in the MTOR-AKT pathway initially identified in tissue-based NGS and confirmed in cfDNA. One patient with a TSC2 splice site mutation had stable disease for 5.4 months, while a second patient with a frameshift deletion had progressive disease after 2.5 months.

Associations of VAF With Clinicopathologic Parameters, Tumor Burden, and Outcomes

Of the variables selected for analysis, higher average VAF was associated with AFP ≥400 ng/mL (β, .38; CI, 0.00 to 0.76; P = .048), liver tumor volume (β, .14; CI, 0.07 to 0.22; P ≤ .001), TTV (β, .15; CI, 0.07 to 0.23; P ≤ .001), and no previous therapy (β, –.5; CI, –0.86 to 013; P = .008; Table 3). In a multivariable model, we observed that patients with AFP ≥400 (β, .43; CI, 0.12 to 0.74; P = .007), higher tumor volume (β, .14; CI, 0.07 to 0.22; P ≤ .001), and no previous therapy (β, –.52; CI, –0.83 to 0.21; P ≤ .001) had higher cfDNA VAF.

TABLE 3.

Univariate and Multivariate Analysis of Clinical Characteristics Associated With Variant Allele Fraction in cfDNA

No mutations were detected by cfDNA analysis in four patients. Among the four patients for whom no alterations were detected in cfDNA, three had an AFP <400 ng/mL and three had low TTV (102 cm³, 20 cm³, and 5 cm³, median in entire cohort 1,221 cm³). Two patients had cfDNA assessed after systemic therapy with one sample collected after a complete radiologic response to immunotherapy.

With a median follow-up time of 19.2 months (1-66.2 months), in those who had cfDNA analyzed before systemic therapy (n = 24), OS was not significantly different in those who had low cfDNA VAF compared with high VAF dichotomized at the median (hazard ratio [HR], 0.88; 95% CI, 0.28 to 2.75; P = .82). Increased VAF was not associated with worse survival outcomes (HR, 1.18; 95% CI, 0.78 to 1.8; P = .43).

DISCUSSION

We demonstrate the feasibility of characterizing the genomic landscape of advanced HCC through targeted NGS of cfDNA. We identify a high rate (92.2%) of detection of tumor-associated alterations in a population with advanced HCC who were not amenable to locoregional therapies, surgical resection, or liver transplant. When compared with previous reports interrogating the utility of cfDNA in solid tumors, the proportion of patients with detected alterations in this study appears notably higher—possibly reflecting the advanced nature of the patient population under study, a potentially higher cfDNA shed rate in patients with HCC, or greater sensitivity of the MSK-ACCESS assay versus the assays used in previous studies.²⁹ Molecular profiling of cfDNA identified frequent mutations in oncogenes shown to be recurrently altered in HCC in previous tumor tissue–based studies, including TERT, CTNNB1, and TP53. Such alterations were observed at similar frequency to what has been reported previously from large, genomic studies of HCC tumor samples.^4,5,30 Furthermore, for genes covered by the MSK-ACCESS assay design, the proportion of alterations detected in our study population was not significantly different than that observed by tumor testing, helping to credential molecular profiling of cfDNA in this disease as a valid clinical approach to molecular genotyping of recurrently mutated genes.

One merit of our approach was that all patients had histologically confirmed HCC, and indeed a considerable proportion of patients in our study had matched tumor tissue sequencing allowing for assessment of plasma-tissue mutational concordance. Importantly, substantial to near-perfect agreement was observed for the most frequently mutated genes (ie, TERT, TP53, and CTNNB1); however, the plasma-based analysis failed to detect all alterations detected in patient-matched tumor samples, while nominating others that were not observed in the paired tumor sample. It is possible that discordant findings were the result of inter- and intra-tumoral genomic heterogeneity that is known to occur in HCC.^31-33 Alternatively, the time between tumor and liquid sampling/analysis (ie, tumor clonal evolution) or differences in the sensitivity of the assay used may have also contributed to the discordant results. Nevertheless, as observed in other solid tumors, the data suggest that tumor and cfDNA analyses are complimentary and that both may be needed in some patients to detect all actionable genomic alterations.^34-36

Our data also suggest that mutational profiling of cfDNA can detect potentially actionable genomic alterations in advanced HCC, and thus might be used in an investigational context to match patients to basket or disease-specific therapeutic clinical trials. Indeed, 4/10 (40%) of alterations detected exclusively by cfDNA profiling were potentially actionable (OncoKB level 3b). This may be of particular utility in those genes where the level of agreement between plasma and tissue-based sequencing was higher and for genes in which molecularly guided clinical trial options are available (TSC2; ClinicalTrials.gov identifier: NCT05103358; CTNNB1 ClinicalTrials.gov identifier: NCT04008797, ClinicalTrials.gov identifier: NCT05091346). Consistent with previous tissue-based assessments where clinical actionability ranged from 24% to 38%, we similarly identified potentially actionable alterations in 37% of patients with HCC.^4,31,37 This is, however, lower than the reported 56.9% in cfDNA samples where the diagnosis was not always made pathologically, emphasizing the importance of studying the clinical utility of targeted gene sequencing in a histologically defined cohort.¹⁷

Our data suggest average VAF may associate with disease burden and other patient-specific features. We aimed to determine what clinical characteristics predict for higher concentrations of circulating cfDNA and what genetic characteristics influence disease outcomes. In univariate analysis, VAF was significantly associated with several markers of disease burden including AFP ≥400, liver tumor volume, and TTV. Importantly, these associations were maintained in a multivariable model, and higher quantities of tumor-derived cfDNA has been previously shown to correlate with stage and overall burden of disease in HCC as well as other solid tumors.^11,38,39 Previous systemic therapy was associated with lower levels of cfDNA VAF suggesting improved yield in treatment-naïve patients.

In HCC patients post-transplant or with resected early-stage disease, it is clear that the detection of tumor-derived cfDNA is highly prognostic of disease recurrence.^40-42 There is, however, little evidence to date in patients with HCC to support the use of cfDNA in terms of guiding treatment selection, monitoring response to therapy, and predicting overall clinical outcomes.⁴³ Although only a small subset of patients in our cohort (n = 24) were treatment-naïve at the time of cfDNA analysis, VAF concentration did not appear to influence survival outcomes; however, large cohorts of patients will be required to assess the prognostic significance of cfDNA detection in patients with advanced HCC.

As with all single-institution studies, there was potential for selection bias. Although MSK-ACCESS samples were collected prospectively, the analysis of clinical characteristics and treatment outcomes and their association with genomic features was retrospective. Timing of cfDNA collection was heterogeneous, with subsequent small sample size for subgroup analysis with further prospective validation required in larger cohorts to determine the impact of cfDNA on clinical outcomes. Nonetheless, this study effectively demonstrates the potential utility of cfDNA analysis in the absence of tissue for identifying oncogenic genomic alterations that are putative, predictive, and prognostic biomarkers that could be used in clinical decision making.

APPENDIX

TABLE A1.

MSK-ACCESS Panel

TABLE A2.

Clinical Characteristics at the Time of Sample Collection Associated With Individual Patients

TABLE A3.

Comparison of Frequencies of Genetic Alteration Identified in Previous Cohort of Tissue-Based Sequencing and cfDNA Sequencing

TABLE A4.

Agreement Between cfDNA and Tumor DNA Mutational Profile

DATA SHARING STATEMENT

All data are available for visualization and analysis via the cBioPortal for Cancer Genomics https://www.cbioportal.org/study/summary?id=hcc_jcopo_msk_2023.

AUTHOR CONTRIBUTIONS

Conception and design: Darren Cowzer, Danny N. Khalil, Imane H. El Dika, David B. Solit, Ghassan K. Abou-Alfa, James J. Harding

Financial support: Darren Cowzer, David B. Solit, Ghassan K. Abou-Alfa, James J. Harding

Administrative support: Darren Cowzer, Joseph S. Light, William Jarnagin, David B. Solit, Ghassan K. Abou-Alfa, James J. Harding

Provision of study materials or patients: Darren Cowzer, Imane H. El Dika, Katrina Columna, Joseph S. Light, Alice C. Wei, David B. Solit, Ghassan K. Abou-Alfa, James J. Harding

Collection and assembly of data: Darren Cowzer, Jessica B White, Pin-Jung Chen, Amin Yaqubie, Joseph S. Light, William Jarnagin, David B. Solit, Ronak H. Shah, Ghassan K. Abou-Alfa, James J. Harding

Data analysis and interpretation: Darren Cowzer, Jessica B White, Joanne F. Chou, Tae-Hyung Kim, Imane H. El Dika, Katrina Columna, Jinru Shia, Hooman Yarmohammadi, Joseph Patrick Erinjeri, Alice C. Wei, William Jarnagin, Richard K. G. Do, David B. Solit, Marinela Capanu, Ronak H. Shah, Michael F. Berger, Ghassan K. Abou-Alfa, James J. Harding

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).