ORIGINAL RESEARCH

Comparative Efficacy of Cabozantiniband Regorafenib for Advanced HepatocellularCarcinoma

Robin K. Kelley . Patrick Mollon . Jean-Frédéric Blanc .

Bruno Daniele . Thomas Yau . Ann-Lii Cheng . Velichka Valcheva .

Florence Marteau . Ines Guerra . Ghassan K. Abou-Alfa

Received: April 29, 2020 / Published online: May 18, 2020� The Author(s) 2020

ABSTRACT

Background: No trials have comparedcabozantinib and regorafenib for the second-line treatment of advanced hepatocellular car-cinoma (HCC).Objectives: Conduct a matching-adjusted indi-rect comparison (MAIC) of the efficacy andsafety of second-line cabozantinib and rego-rafenib in patients with advanced HCC anddisease progression after prior sorafenib.

Methods: The CELESTIAL and RESORCE trialswere used for indirect comparison of second-line cabozantinib and regorafenib in advancedHCC. Population-level data were available forRESORCE, individual patient data (IPD) forCELESTIAL. To align with RESORCE, theCELESTIAL population was limited to patientswho received first-line sorafenib only. To mini-mize potential effect-modifying populationdifferences, the CELESTIAL IPD were weightedto balance the distribution of clinically relevantbaseline characteristics with those of RESORCE.Overall survival (OS) and progression-free sur-vival (PFS) were evaluated for the matching-adjusted second-line CELESTIAL populationand compared with those for RESORCE usingweighted Kaplan-Meier curves and parametricmodeling. Rates of grade 3/4 treatment-

Digital Features To view digital features for this articlego to https://doi.org/10.6084/m9.figshare.12213854.

Electronic Supplementary Material The onlineversion of this article (https://doi.org/10.1007/s12325-020-01378-y) contains supplementary material, which isavailable to authorized users.

R. K. Kelley (&)UCSF, Helen Diller Family Comprehensive CancerCenter, San Francisco, CA, USAe-mail: katie.kelley@ucsf.edu

P. Mollon � V. Valcheva � F. MarteauIpsen Pharma, Boulogne-Billancourt, France

J.-F. BlancHôpital Haut-Lévêque, CHU de Bordeaux, Bordeaux,France

B. DanieleAzienda Ospedaliera G Rummo, Benevento,Italy

B. DanieleOspedale del Mare, Naples, Italy

T. YauUniversity of Hong Kong, Pokfulam, Hong Kong

A.-L. ChengNational Taiwan University Cancer Center, NationalTaiwan University Hospital, Taipei, Taiwan,Republic of China

I. GuerraIQVIA Ltd, London, UK

G. K. Abou-AlfaMemorial Sloan Kettering Cancer Center, New York,NY, USA

G. K. Abou-AlfaWeill Medical College at Cornell University, NewYork, NY, USA

Adv Ther (2020) 37:2678–2695

https://doi.org/10.1007/s12325-020-01378-y

https://orcid.org/0000-0002-1984-2430https://orcid.org/0000-0001-6172-9485https://orcid.org/0000-0002-5221-9755https://orcid.org/0000-0002-9152-6512https://orcid.org/0000-0002-3476-0046https://orcid.org/0000-0002-1522-8054https://doi.org/10.6084/m9.figshare.12213854https://doi.org/10.1007/s12325-020-01378-yhttps://doi.org/10.1007/s12325-020-01378-yhttps://doi.org/10.1007/s12325-020-01378-yhttps://doi.org/10.1007/s12325-020-01378-yhttp://crossmark.crossref.org/dialog/?doi=10.1007/s12325-020-01378-y&domain=pdfhttps://doi.org/10.1007/s12325-020-01378-y

emergent adverse events (TEAEs) affecting[ 5%of patients in any study arm were compared.Results: In the matching-adjusted second-linepopulations (CELESTIAL, effective samplesize = 266; RESORCE, n = 573), median (95%confidence interval) OS was similar forcabozantinib and regorafenib (11.4 [8.9–17.0]versus 10.6 [9.1–12.1] months; p = 0.3474, log-rank test). Median PFS was longer for cabozan-tinib than regorafenib (5.6 [4.9–7.3] versus 3.1[2.8–4.2] months; p = 0.0005, log-rank test).There was a trend for lower rates of some grade3/4 TEAEs with regorafenib than withcabozantinib, which may reflect the exclusionof sorafenib-intolerant patients from RESORCEbut not from CELESTIAL, a difference that theMAIC methods could not remove. Only diar-rhea rates were statistically significantly lowerfor regorafenib (p B 0.001).Conclusions: Cabozantinib may achieve simi-lar OS and prolonged PFS compared with rego-rafenib in patients with progressive advancedHCC after prior sorafenib.

PLAIN LANGUAGE SUMMARY

Cabozantinib and regorafenib are treatmentsapproved for some patients with advancedhepatocellular carcinoma (HCC), a type of livercancer, after disease progression despite priorsorafenib treatment. Cabozantinib, regorafeniband sorafenib are tyrosine kinase inhibitors(TKIs), meaning that they slow cancer progres-sion by targeting specific ways that tumorsgrow. Cabozantinib and regorafenib offer ben-efits to patients compared with placebo (i.e., notreatment) for those who have progressed

despite sorafenib treatment. No clinical studieshave compared cabozantinib and regorafenibdirectly. This study compared the efficacy andsafety of cabozantinib and regorafenib usingdata from trials of each drug versus placebo:CELESTIAL for cabozantinib and RESORCE forregorafenib. These two trials were similar—bothinvolved patients with progressive advancedHCC who had received previous cancer treat-ment. There were some important differences,but these were minimized using statisticalmethods (matching and adjustments/‘‘weight-ing’’) allowing outcomes to be meaningfullycompared. One difference that could not beremoved by the statistical methods was thatpatients who were intolerant to prior sorafenibwere excluded from RESORCE but were eligiblefor the CELESTIAL trial. In the otherwise mat-ched populations, treatment with cabozantinibwas associated with similar overall survival andsignificantly longer progression-free survivalthan regorafenib. Rates of diarrhea were signif-icantly lower for regorafenib than cabozantinib,suggesting that regorafenib may be better tol-erated, but this may reflect the exclusion ofsorafenib-intolerant patients from RESORCE.These findings cannot replace a head-to-headstudy, but may help in guiding decision-makingbetween cabozantinib and regorafenib inpatients with progressive advanced HCC aftersoraftenib treatment.

Keywords: Cabozantinib; CELESTIAL; Hepato-cellular carcinoma (HCC); Indirect treatmentcomparison; Matching-adjusted indirectcomparison (MAIC); Regorafenib; RESORCE;Second-line; Systemic therapy; Targeted therapy

Adv Ther (2020) 37:2678–2695 2679

Key Summary Points

Why carry out this study?

The tyrosine kinase inhibitorscabozantinib and regorafenib areapproved for the treatment of patientswith advanced hepatocellular carcinoma(HCC) who have progressed despite priortreatment with sorafenib.

No clinical trials have directly comparedcabozantinib and regorafenib for thesecond-line treatment of advanced HCC.

This matching-adjusted indirectcomparison (MAIC) used data from thephase 3 CELESTIAL (NCT01908426) andRESORCE (NCT01774344) trials togenerate comparative efficacy and safetyestimates for cabozantinib versusregorafenib in patients with progressiveadvanced HCC after prior sorafenibtherapy.

What was learned from this study?

In patients with progressive advancedHCC who have received prior sorafenibtreatment, cabozantinib may achievesimilar overall survival and prolongprogression-free survival compared withregorafenib; regorafenib may be associatedwith lower rates of grade 3 or 4 diarrhea.

A MAIC cannot replace a head-to-headrandomized controlled trial, but thesefindings may help in guiding clinicaldecision-making between cabozantiniband regorafenib when treating patientswith progressive advanced HCC in theabsence of direct trial evidence.

INTRODUCTION

The era of targeted therapy for liver cancerbegan in 2007 with the approval of sorafenib forthe first-line management of patients with

advanced hepatocellular carcinoma (HCC)[1–4]. More than a decade later, the range oftherapeutic options for HCC has broadened toinclude a number of new targeted therapieswith proven survival benefit in phase 3 trials. Inthe first-line setting, the tyrosine kinase inhi-bitor (TKI) lenvatinib has now been approvedfor patients with unresectable disease [5, 6].Approved second-line treatment options (afterprior treatment with sorafenib) now include theTKIs regorafenib [7, 8] and cabozantinib [9, 10]in patients with advanced HCC, the anti-vascular endothelial growth factor receptor 2(VEGFR2) monoclonal antibody ramucirumabin patients with alpha-fetoprotein levels[ 400 ng/ml [11, 12] and the checkpoint inhi-bitors nivolumab and pembrolizumab (pro-grammed cell death receptor-1 antibodies) asmonotherapy or, in the case of nivolumab, incombination with ipilimumab (a cytotoxicT-lymphocyte-associated protein 4 antibody)[13–16].

The arrival of second-line agents foradvanced HCC extends the previous therapeu-tic offering for patients, many of whom presentwith advanced disease at the time of diagnosisand have had few therapeutic options availableto them [17]. Accordingly, 5-year survival ratesfor localized, regional and distant HCC aretypically poor (31%, 11% and 2%, respectively)[18], highlighting the importance of newsecond-line therapies and the clinical imperativeto optimize their use.

The TKIs regorafenib and cabozantinib areboth approved as second-line agents forpatients with HCC after prior treatment withsorafenib [7–10]. They share a drug class andoral mode of administration, but they differ intheir molecular targeting profiles. Regorafenibtargets multiple receptor tyrosine kinases,including those involved in tumor angiogenesis(VEGFR-1, -2, -3, TIE2), oncogenesis (KIT, RET,RAF-1, BRAF, BRAFV600E), metastasis (VEGFR3,PDGFR, FGFR) and tumor immunity (CSF1R)[19]. Cabozantinib also has inhibitory activityagainst tumor angiogenesis and oncogenesis,but it additionally targets the hepatocytegrowth factor receptor protein (MET), involvedin tumor growth and invasion, and othertyrosine kinases, including those involved in

2680 Adv Ther (2020) 37:2678–2695

modulation of tumor immunity (e.g., AXL,MER) [20, 21]. Current HCC managementguidelines recommend either agent in advanceddisease following progression after sorafenibon the basis of evidence from the pivotalRESORCE (NCT01774344) and CELESTIAL(NCT01908426) phase 3 trials [1, 8, 10, 22, 23].

RESORCE compared once-daily regorafenib160 mg to placebo for weeks 1–3 of every 4-weekcycle (in patients with HCC who were tolerantto sorafenib, but who had progressed duringsorafenib treatment; n = 573) [8]. CELESTIALcompared once-daily cabozantinib 60 mg toplacebo (continuous dosing) in patients withadvanced HCC who had progressed after at leastone systemic treatment for HCC and may havereceived up to two previous systemic regimensfor advanced HCC (n = 707), one of which wasrequired to be sorafenib [10].

In the second-line (sorafenib-tolerant) pop-ulation included in RESORCE, regorafenibimproved median overall survival (OS) andprogression-free survival (PFS) compared withplacebo (hazard ratio [HR] [95% confidenceinterval, CI], 0.63 [0.50–0.79] and 0.46[0.37–0.56], respectively; one-sided p\0.0001for both) [8]. The most common clinicallyrelevant grade 3 or 4 treatment-emergent adverseevents (TEAEs) were hypertension, hand-footskin reaction, fatigue and diarrhea [8]. In themixed second- and third-line populationincluded in CELESTIAL, cabozantinib alsosignificantly improved OS and PFS comparedwith placebo (HR [95% CI], 0.76 [0.63–0.92],two-sided p = 0.005 and 0.44 [0.36–0.52], two-sided p\0.001, respectively). Cabozantinibalso prolonged median OS and PFS in thesubgroup of patients who had only receivedprevious systemic treatment with sorafenib(stratified HR [95% CI], 0.70 [0.55–0.88] and0.40 [0.32–0.50], respectively). Similar toRESORCE, hand-foot reaction (assessed aspalmar-plantar erythrodysesthesia), fatigue anddiarrhea were among the most common grade 3or 4 events; higher rates of hypertension andincreased aspartate aminotransferase (AST) levelwere also recorded with cabozantinib comparedwith placebo [10].

While the CELESTIAL and RESORCE trialsprovide robust evidence for the second-line use

of either cabozantinib or regorafenib in pro-gressive advanced HCC, at the time of writing,there have been no randomized controlled trialsto compare the two agents directly. In theabsence of head-to-head trial data, indirecttreatment comparisons offer a means ofestimating probable treatment outcomes ifcomparator drugs were used in the same way insimilar patients [24, 25]. Standard indirecttreatment comparisons estimate the effect sizesof individual treatments relative to a commonreference arm (e.g., placebo); these relativeestimates are then compared. This approach isused in network meta-analyses, but it relies onthe studies involved being sufficiently similar inall respects other than the treatments beingcompared [26]. When this is not the case and astandard indirect treatment comparison is notvalid, population-adjustment methods are usedto minimize between-trial population differ-ences. These approaches weight individualpatient data (IPD) to reduce differences in thedistribution of clinically relevant covariates,thereby reducing the potential for bias incomparative outcome estimates. Propensityscore matching is one method of populationmatching, but it requires IPD to be available forall studies being compared [25]. When IPD areavailable for only one of two studies beingcompared, a matching-adjusted indirectcomparison (MAIC) method can be used. A MAICweights the IPD for the available study so thatits baseline characteristics match those of areference comparator study for which onlypublished aggregate-level data are available [25].MAIC analyses are routinely used for healthtechnology assessment [24, 25], and theirpotential to inform clinical decision-making inthe absence of direct comparative data has beenutilized across a range of cancer types (e.g.,breast cancer, prostate cancer, basal cell carci-noma) [27–29], including HCC [30–32].

We report here the first MAIC of second-lineTKI options for HCC, providing an assessmentof the comparative efficacy and safety ofcabozantinib and regorafenib for patients withadvanced HCC who have received sorafenib asthe only prior systemic therapy (see Fig. 1 for agraphical summary of the study).

Adv Ther (2020) 37:2678–2695 2681

METHODS

Data Source Identification And Eligibility

The CELESTIAL trial of cabozantinib (n = 707)[10] and RESORCE trial of regorafenib (n = 573)[8] were identified in the published literatureand assessed for their feasibility for a standardindirect treatment comparison (ITC) of TKIsrecently approved for the second-line manage-ment of advanced progressive HCC followingprior sorafenib treatment.

RESORCE included a second-line populationof patients with advanced HCC who toleratedsorafenib, but who had progressed followingsorafenib treatment only [8, 12]. In contrast,CELESTIAL included a mixed second- and third-line population [10]. While CELESTIAL patientsmust have received previous treatment withsorafenib (by default), they may also havereceived a second systemic treatment foradvanced HCC prior to recruitment to the trial.Furthermore, CELESTIAL did not excludepatients who were intolerant to sorafenib.

Despite these notable population differences,the availability of IPD for CELESTIAL (obtainedfrom Exelixis with a data cutoff date of 1 June2017) enabled isolation of a pure second-lineCELESTIAL subpopulation for comparison withRESORCE. The similarity of the efficacy andsafety outcomes reported for the two trialsconfirmed their potential for use in an indirecttreatment comparison of second-line cabozan-tinib versus regorafenib in patients with pro-gressive advanced HCC.

To evaluate whether a standard ITC wasfeasible for RESORCE (regorafenib) and thesecond-line CELESTIAL (cabozantinib) subpop-ulation, a panel of expert oncologists convenedin June 2018 to review the design similaritiesand baseline population characteristics of thetwo trials (Supplemental Tables S1 and S2). Thepanel identified differences in several effect-modifying covariates with the potential to biasa standard ITC. Therefore, a MAIC analysisapproach was selected to compare the safetyand efficacy of regorafenib, using the aggregatepopulation data published for RESORCE and

Fig. 1 Overview of the study

2682 Adv Ther (2020) 37:2678–2695

IPD for the second-line subpopulation fromCELESTIAL.

Matching-Adjusted Indirect Comparison:Cabozantinib and Regorafenib

The MAIC analysis was conducted in accor-dance with the established methodology out-lined in the guidance from the NationalInstitute for Health and Care Excellence [25, 33]as summarized in Supplemental Fig. S1 (see theelectronic supplementary material).

Population MatchingThe IPD from CELESTIAL were used to identify apure second-line population of patients whoreceived cabozantinib after sorafenib only. TheIPD also allowed the clinically meaningful differ-ences in potential effect-modifying covariates tobe minimized by weighting the baseline charac-teristics of the second-line CELESTIAL subpopu-lation so as to match them to those of thepublished RESORCE population-level statistics(e.g., means, medians, percentages). The baselinedifferences between the trials that had beenidentified as being potential effect modifiers bythe expert panel were assessed for colinearity, andthe following covariates were selected as thematching criteria: age, race, geographical region,Eastern Cooperative Oncology Group Perfor-mance Status, Child-Pugh class, duration of priorsorafenib treatment, extrahepatic disease,macrovascular invasion, etiology of HCC (hep-atitis B, alcohol use and hepatitis C) and serumalpha-fetoprotein level. Patients in CELESTIALwith missing IPD for any of the selected matchingcriteria were excluded from the analysis.

Outcome Evaluation: Survival AnalysisSurvival outcomes for the matching-adjustedsecond-line CELESTIAL population were thenevaluated and compared with those published forRESORCE. For cabozantinib, median OS and PFSestimates were derived from weighted Kaplan-Meier (KM) curves fitted to the survival data [34].For cabozantinib, confidence intervals (CIs) forthe KM weighted curves were generated fromsimulations [35] and for median survival usingWoodruff’s method [36]. For regorafenib, KM

curves and median survival estimates weresourced from the RESORCE publication [8].

MAIC analyses can be ‘anchored’ or‘unanchored,’ with methodologic guidelinesgenerally favoring an anchored analysis approach,when feasible [25]. An anchored MAIC gener-ates estimates relative to a common comparatorarm, such as placebo. An unanchored analysisinvolves no common comparator arm and basesestimates on absolute outcomes. An unan-chored analysis involves more assumptionsthan an anchored approach, but is necessarywhen there is no common comparator arm(e.g., single-arm studies) or when assumptionsunderpinning an anchored approach are notsatisfied. For anchored hazard ratios to be valid,the proportional hazards assumption must besatisfied (i.e., the treatment effect must be pro-portional over time and the survival curves fit-ted to each treatment group must have a similarshape) [25, 33].

In the present analysis, the feasibility ofconducting an anchored analysis (with placeboas the common comparator arm) was assessed.The proportional hazards assumption was tes-ted for OS and PFS by visual inspection of thelog-cumulative hazard plots to ensure that therewas no pattern of non-parallelism. The test wasconducted for the matching-adjusted andunmatched second-line cabozantinib popula-tions versus placebo from CELESTIAL, for theregorafenib versus placebo populations fromRESORCE and for the matching-adjustedand unmatched second-line cabozantinibCELESTIAL population versus the regorafenibRESORCE population. The findings of the visualinspection of the log-cumulative hazard plotswere then validated by visual inspection of thescaled Schoenfeld residuals and using theGrambsch-Therneau test (a statistical test basedon the scaled Schoenfeld residuals) [33, 37].

Where the proportional hazards assumptionwas not satisfied and an anchored analysis notsupported, an unanchored analysis was con-ducted by fitting individual parametric survivalcurves to each treatment arm, in line with bestpractice guidelines. Parametric model selectionwas based on an analysis of Akaike’s informa-tion criterion and Shwarz’s Bayesian

Adv Ther (2020) 37:2678–2695 2683

information criterion (AIC/BIC), with superiormodel fit indicated by lower AIC and BIC [33].

Outcome Evaluation: SafetyIncidence of grade 3 or 4 TEAEs affecting[ 5%of patients in any of the second-line CELESTIALor RESORCE treatment arms was compared forcabozantinib versus regorafenib. In an anchoredanalysis, the estimated relative effects (log oddsratios [ORs]) of cabozantinib versus placebo inthe weighted population were generated andcompared with log-ORs for regorafenib versusplacebo, computed from the published data.Safety outcome estimates of cabozantinib versusregorafenib were constructed in the log-ORscale.

If a TEAE of interest did not occur in either ofthe placebo arms of the trials (preventing ananchored log-OR analysis), an unanchoredanalysis of the active treatment arms was con-ducted. The number of TEAEs occurring inCELESTIAL was used to compute a weighted,unanchored estimate.

Analyses

The analyses were performed using R version3.5.2 (R Core Team, 2014). The package ‘survey’version 3.36 was used to fit weighted survivalmodels with weights computed from the MAICused as sampling weights.

Compliance with Ethics Guidelines

The results presented in this manuscript arebased on previously published studies. All pro-cedures performed in those studies involvinghuman participants were in accordance withthe ethical standards of the local InstitutionalReview Boards for each site and with the 1964Helsinki Declaration and its later amendmentsor comparable ethical standards. Informedconsent was obtained from all individual par-ticipants included in the CELESTIAL (Clini-calTrials.gov identifier NCT01908426) andRESORCE (ClinicalTrials.gov identifierNCT01774344) trials.

RESULTS

Patient Characteristics

RESORCE involved a total of 573 second-linepatients with advanced HCC who were ran-domized to regorafenib (n = 379) or placebo(n = 194) following progression on sorafenib. Intotal, 495 second-line patients were enrolled inCELESTIAL and randomized to cabozantinib(n = 331) or placebo (n = 164). When limited tothe CELESTIAL patients for whom data wereavailable for all effect-modifying characteristics,the pre-matched population size reduced fur-ther to 484 patients (cabozantinib, n = 326;placebo, n = 158) and, following matching, toan effective sample size of 266 patients (non-additive with respect to each treatment arm:cabozantinib, n = 187; placebo, n = 81)(Table 1).

Application of MAIC weighting to the base-line IPD from CELESTIAL was effective in bal-ancing the effect-modifying baselinecharacteristics of the second-line CELESTIALand RESORCE populations, most notablypatient ethnicity and geographical region oforigin, HCC etiology, proportion of patientswith Eastern Cooperative Oncology Group(ECOG) performance status 0 and HCC etiology(Table 2). As a result, there were no discernableclinically relevant differences between thebaseline characteristics of the matching-ad-justed second-line CELESTIAL population andthe RESORCE populations (Table 2).

Survival Outcomes

KM-Derived EstimatesEstimated median (95% CI) OS derived fromthe weighted KM curves were similar:11.4 (8.9–17.0) months for the matching-adjustedcabozantinib population and 10.6 (9.1–12.1)months for the regorafenib population(p = 0.3474, log-rank test). In comparison,equivalent OS estimates for the placebo armswere 7.2 (6.1–10.8) for the matching-adjustedsecond-line CELESTIAL population and7.8 (6.3–8.8) months for RESORCE (Fig. 2 andTable 3).

2684 Adv Ther (2020) 37:2678–2695

Estimated median (95% CI) PFS was5.6 (4.9–7.3) months for the matching-adjustedcabozantinib population compared with3.1 (2.8–4.2) months for regorafenib, a statisticallysignificant difference (p = 0.0005, log-rank test).For the placebo arms, equivalent estimates were1.9 (1.9–2.1) for the matching-adjusted second-line CELESTIAL population and 1.5 (1.4–1.6)months for RESORCE (Fig. 3 and Table 3).

Parametric Modeling EstimatesVisual inspection of the log of cumulative haz-ard versus time plots displayed a distinct patternof non-parallelism for both OS and PFS, indi-cating that the proportional hazards assump-tion was not valid, and an anchored analysiswas not supported (Supplemental Fig. S2). Theplots of scaled Schoenfeld residuals versus timeshowed a systematic departure from the hori-zontal for both outcomes, confirming theassessment that the proportional hazardsassumption was not supported, as did the non-zero slopes (at the 5% significance level) givenby the Grambsch-Therneau test (SupplementalFig. S3). In line with recommended practice, anunanchored analysis was therefore conducted

by fitting individual parametric survival curvesto each treatment arm [33].

Analysis of AIC and BIC for differentcandidate models identified the log-logisticdistribution as the best-fit model for OS and thegeneralized gamma model as the best-fit modelfor PFS (Supplemental Tables S3 and S4). Thedirection of the survival trends generated by theparametric modeling mirrored those of theweighted KM analyses for both the activetreatment and the placebo arms (SupplementalTable S5). As for the KM-derived estimates,median (95% CI) OS estimates were similar andthe 95% CIs overlapped: 11.40 (10.01–12.96)months for the matching-adjusted second-line cabozantinib population versus10.29 (9.15–11.56) months for the regorafenibpopulation. For the placebo arms equivalentestimates were 8.27 (7.00–9.76) months for thematching-adjusted second-line CELESTIALpopulation and 7.30 (6.30–8.47) months forRESORCE (Table S5).

For PFS, the median (95% CI) estimate waslonger for the matching-adjusted second-linecabozantinib population (5.49 [4.92–6.13]months) compared with the regorafenib popu-lation (3.39 [3.05–3.78] months). There was no

Table 1 MAIC population sizes

CELESTIAL population RESORCEpopulationUnmatched Matching-adjusted

n(overall)

n(second line)

n(second line withnon-missing data)

Effective sample size n (published)

Active

treatment

470 331 326 187 379

Placebo 237 164 158 81 194

Total 707 495 484 266a 573

a Non-additive with respect to each treatment arm. The ESS is computed separately for each patient group asðPi

ŵiÞ2Piðŵ2i Þ(the

squared sum of weights [numerator] and sum of squared weights [denominator]). The ESS computed for the full populationwill not equate to the sum of the ESS for each patient group individually, unless all the weights are the same for eachtreatment armn number of patients enrolled and randomized

Adv Ther (2020) 37:2678–2695 2685

overlap in the CIs for the cabozantinib andregorafenib PFS estimates, suggesting the dif-ference in favor of cabozantinib was statisticallysignificant. For the placebo arms, equivalentestimates were 2.35 (2.11–2.61) months for thematching-adjusted second-line CELESTIALpopulation and 1.87 (1.68–2.09) months forRESORCE (Table S3).

Safety Outcomes

TEAEs of interest (grade 3 or 4 occurring in morethan 5% of patients in any trial arm) were ASTincrease (‘‘increased AST’’ in RESORCE),diarrhea, elevated bilirubin, fatigue, hyperten-sion and palmar-plantar erythrodysesthesiasyndrome (hand-foot skin reaction in RESORCE).

An anchored log-OR analysis found nosignificant difference between the matching-adjusted cabozantinib and regorafenibpopulations in terms of frequency of grade 3 or4 TEAEs: fatigue (p = 0.9313); elevated bilirubin(p = 0.8558) or increased AST (p = 0.2201). Thisresult was consistent for both the matching-adjusted and unmatched cabozantinib popula-tions (Supplemental Table S6). Although therewas a possible trend toward a higher rate ofgrade 3 or 4 hypertension in the matching-adjusted cabozantinib population comparedwith the regorafenib population, the confidenceintervals for the estimate crossed zero (indicat-ing no difference), and a test of the nullhypothesis confirmed that the difference wasnot statistically significant (p = 0.0611) (Fig. 4and Supplemental Table S6).

Table 2 Baseline characteristics of the MAIC populations

CELESTIAL second-line population RESORCE population

Unmatched(n = 495)a

Matching-adjusted(n = 266)b

Published(n = 573)

Age\ 65 years, % 53.33 54.97 54.97

Female, % 17.58 18.63 12.04

Asian geographical region, % 22.83 37.70 37.70

White, % 58.18 35.95 35.95

ECOG performance status 0, % 56.97 65.79 65.79

Child–Pugh class A, % 98.79 97.91 97.91

Duration of prior sorafenib treatment,

mean (months)

7.65 11.63 11.63

Extrahepatic disease, % 76.16 71.90 71.90

Macrovascular invasion, % 29.41 28.62 28.62

Etiology, %

Hepatitis B 37.37 37.70 37.70

Alcohol use 21.52 25.31 25.31

Hepatitis C 25.10 20.77 20.77

Alpha-fetoprotein[ 400 ng/ml, % 40.81 43.46 43.46a Includes patients with missing data for effect-modifying baseline characteristicsb Effective sample size; excludes patients with missing data for effect-modifying baseline characteristicsECOG Eastern Cooperative Oncology Group

2686 Adv Ther (2020) 37:2678–2695

Owing to the low frequency of grade 3 or 4palmar-plantar erythrodysesthesia in theplacebo-treated patients (one case in RESORCE;no cases in CELESTIAL) and no cases of grade 3or 4 diarrhea in the RESORCE placebo arm, ameaningful anchored log-OR estimate couldnot be constructed for these TEAEs.Unanchored estimates were thereforecomputed. The unanchored analysis found nosignificant difference in rates of palmar-plantar erythrodysesthesia between the matching-adjusted cabozantinib and regorafenib popula-tions (p = 0.848), but significantly lower rates ofdiarrhea in the regorafenib population com-pared with both the unmatched and matching-adjusted cabozantinib populations (p = 0.001and p\0.001 respectively) (Fig. 4 and Supple-mental Table S6).

DISCUSSION

Outcome Interpretation

From the present analysis, we report compara-tive efficacy and safety estimates for cabozan-tinib and regorafenib for the second-linetreatment of patients with advanced HCC after

prior sorafenib, using data from the phase 3CELESTIAL and RESORCE trials. The treatmentlandscape for HCC is expanding rapidly, andthere is a resultant need for head-to-head clini-cal trial data to guide second-line HCC treat-ment decisions. In this setting, indirecttreatment comparisons offer standardizedmethods for generating comparative estimatesthat are widely accepted for health technologyassessment [24, 25] and are increasingly recog-nized in the clinical sphere for their potential toguide clinicians in their decision-making[27–32].

A standard indirect treatment comparison ofcabozantinib and regorafenib is not feasibleowing to clinically relevant differences in thebaseline characteristics of the RESORCE popu-lation and the second-line subpopulation fromCELESTIAL. Therefore, a MAIC was selected as amore robust method of comparison, with anunanchored approach selected for the survivalanalysis on the basis of the results of three testsof the proportional hazards (PH) assumption,which indicated that the PH assumption wasnot satisfied. For OS, this conclusion was furthervalidated by the identification of the log-logisticmodel as the best-fit parametric model; log-logistic models are accelerated failure time models

0 5 10 15 20 25 30 35

64

326 221 140 87 51 27 16158 87 45 24 16 11 7

Months

0.00

0.25

0.50

0.75

1.00(a)

Prob

abilit

y of

sur

viva

l

Number of patients at risk

CabozantinibPlacebo95% CI

0 5 10 15 20 25 30 35

379 247 153 78 41 13194 111 49 26 11 5

Months

0.00

0.25

0.50

0.75

1.00(b)

Prob

abilit

y of

sur

viva

l

Number of patients at risk

RegorafenibPlacebo

CabozantinibPlacebo

RegorafenibPlacebo

95% CI

Fig. 2 Kaplan-Meier curves for overall survival in the matching-adjusted second-line CELESTIAL population (a) and theRESORCE population (b). CI confidence interval

Adv Ther (2020) 37:2678–2695 2687

and do not produce a single hazard ratio, mak-ing them incompatible with the PHassumption.

The matching and statistical adjustmentsteps of a MAIC analysis help to reduce thepotential for bias by minimizing populationdifferences that might modify or obscure theability to discern true treatment effects. In thepresent analysis, weighting the baseline IPD forsecond-line CELESTIAL patients to align themwith the baseline RESORCE characteristics waseffective in reducing discernible populationdifferences. In the unanchored survival analy-sis, cabozantinib was associated with similar OSand prolonged PFS compared with regorafenib.These findings were consistent across analysesfor both the KM-derived and parametric mod-eling survival estimates. There is, however, aneed for caution when interpreting the PFSresult owing to differences in the tumor assess-ment schedules used in the CELESTIAL andRESORCE trials. RESORCE assessed tumorgrowth every 6 weeks for the first eight cyclesand every 12 weeks thereafter during treatment.

In CELESTIAL, tumors were assessed every8 weeks after randomization and performeduntil 8 weeks after radiographic progression ortreatment/placebo discontinuation, whicheveroccurred later. Assessment of PFS was, therefore,initially more frequent for patients in RESORCEthan in CELESTIAL and, thereafter, more fre-quent in CELESTIAL than in RESORCE. Thisdifference may have introduced bias into thePFS result in the current analysis. At the indi-vidual patient level, the direction of any suchbias would depend on the timing of tumorgrowth. It would, for example, favor cabozan-tinib if tumor growth occurred at week 10(assessed at week 12 in RESORCE, but not untilweek 16 in CELESTIAL), yet favor regorafenib iftumor growth occurred at week 13 (assessed atweek 16 in CELESTIAL, but not until week 18 inRESORCE). Thus, the overall direction of bias, ifany, remains unclear.

The grade 3 or 4 TEAE profiles for regorafeniband cabozantinib were generally similar forboth matching-adjusted and unmatchedcabozantinib populations (Fig. 4). The

Table 3 Median survival estimates for the matching-adjusted second-line CELESTIAL population and the RESORCEpopulation: weighted Kaplan-Meier estimates

KM-derived estimate, months(median [95% CI])

p value

Overall survival

Active treatment Cabozantinib (ESS = 187) 11.4 (8.9–17.0) 0.3474a

Regorafenib (n = 379) 10.6 (9.1–12.1)

Placebo CELESTIAL (ESS = 81) 7.2 (6.1–10.8) NE

RESORCE (n = 194) 7.8 (6.3–8.8)

Progression-free survival

Active treatment Cabozantinib (ESS = 187) 5.6 (4.9–7.3) 0.0005a

Regorafenib (n = 379) 3.1 (2.8–4.2)

Placebo CELESTIAL (ESS = 81) 1.9 (1.9–2.1) NE

RESORCE (n = 194) 1.5 (1.4–1.6)

CI confidence interval, ESS effective sample size, KM Kaplan-Meier, NE not evaluateda Log-rank test

2688 Adv Ther (2020) 37:2678–2695

exception was rates of grade 3 or 4 diarrhea,which were statistically significantly lower inthe regorafenib population than in either theunmatched or matching-adjusted cabozantinibpopulations. There was also a trend towardhigher rates of hypertension in the cabozan-tinib compared with regorafenib populations;however, the difference was not statisticallysignificant. When interpreting the safetyresults, it is also noteworthy that sorafenib-intolerant patients were excluded from RESORCE,but not from CELESTIAL. The MAIC procedureswere not able to adjust for this between-trialdifference, resulting in the possibility of a biastoward higher rates of TKI treatment intoler-ance (and possibly later-stage disease) in theCELESTIAL population. Overall, the wide con-fidence intervals depicted in the forest plot oflog-OR estimates for TEAEs with cabozantinib(versus regorafenib) indicate a high degree ofimprecision in the estimates, likely arising fromthe very low frequency of grade 3 or 4 events inthe placebo arms, particularly in CELESTIAL.

Limitations

Specific limitations associated with the PFS andTEAE estimates have been discussed. While

MAIC procedures can reduce the impact ofpotentially effect-modifying baseline character-istics for reported covariates, they were not ableto adjust for between-trial differences in assess-ment schedules or for the presence of sorafenib-intolerant patients in the CELESTIAL popula-tion (versus their exclusion from RESORCE).Such differences are unavoidable features ofsome indirect treatment comparisons [27, 28]and network meta-analyses [38, 39], but arerelevant factors to consider when interpretingtheir results.

There are additional limitations to MAICanalyses that are also worth of consideration.Matching cannot account for all differencesbetween trial populations, and it is possible thatthe results of this MAIC are affected by someresidual between-trial differences, as evidencedby the difference in survival outcomes for theplacebo arms despite matching and adjustment(Tables 3 and S5). Additional effect modifierscan exist between comparator trials, despitebaseline weighting and matching. For example,treatment adherence following randomizationcan differ in comparator trials, affecting resul-tant drug exposure and influencing treatment-related outcomes. There may also be differencesin unknown prognostic variables (i.e., covari-ates that affect outcome but do not alter

0 5 10 15 20 25 30 35

326 133 48 13 4 2 2158 23 8 2 2 1 0

Months

0.00

0.25

0.50

0.75

1.00(a)Pr

obab

ility

of s

urvi

val

0 5 10 15 20 25 30 35

379 110 34 14 8 4194 20 5 1 0 0

Months

0.00

0.25

0.50

0.75

1.00(b)

Prob

abilit

y of

sur

viva

l

Number of patients at risk

CabozantinibPlacebo95% CI

Number of patients at risk

RegorafenibPlacebo

CabozantinibPlacebo

RegorafenibPlacebo

95% CI

Fig. 3 Kaplan-Meier curves for progression-free survival in the matching-adjusted second-line CELESTIAL population(a) and the RESORCE population (b). CI confidence interval

Adv Ther (2020) 37:2678–2695 2689

treatment effect). In an anchored analysis,purely prognostic variables do not affect inter-pretation of the results because they do notaffect the relative treatment effects for eachdrug versus placebo because of within-studyrandomization. However, in the present unan-chored analysis, where the comparison ofcabozantinib and regorafenib is based on abso-lute rather than relative treatment outcomes,the results may be influenced by potentialimbalances in unknown prognostic variables atstudy entry or (particularly relevant to OS) byheterogeneity in post-progression treatments.Heterogeneity in use of downstream therapies isanother possible cofounding variable. InCELESTIAL, more than one-quarter of patients(26% cabozantinib; 33% placebo) received sub-sequent systemic or local liver-directed anti-cancer therapy [10]. The equivalent proportion

is not reported for RESORCE and is not knownfor the matching-adjusted second-line CELES-TIAL population, but some degree of between-trial heterogeneity in use of post-progressiontreatments is likely.

Finally, the reduced effective sample sizeresulting from MAIC matching and adjustmentsdecreases the statistical power of subsequentanalyses. A lack of power in this analysis wasreflected in the low rates of TEAE, and theresultant imprecision was evident in the highanchored log-ORs and large CIs for some TEAEs.Furthermore, the necessary use of an unan-chored survival analysis and related discardingof placebo data can result in artificially narrowinterval estimates, with arbitrary implicationson statistical significance. For these reasons, theresults of a MAIC cannot replace evidence froma randomized controlled trial.

Increased aspartate aminotransferaseUnmatched cabozantinib populationMatching-adjusted cabozantinib population

Diarrheaa

Unmatched cabozantinib populationMatching-adjusted cabozantinib population

Elevated bilirubineUnmatched cabozantinib populationMatching-adjusted cabozantinib population

FatigueUnmatched cabozantinib populationMatching-adjusted cabozantinib population

Palmar-plantar erythrodysesthesiaa Unmatched cabozantinib populationMatching-adjusted cabozantinib population

HypertensionUnmatched cabozantinib populationMatching-adjusted cabozantinib population

Favors cabozantinib Favors regorafenib

Log OR estimates422– 0–3 1–4– 1 3 5

Fig. 4 Forest plot of grade 3 or 4 TEAE log-OR (95% CI)estimates for the unmatched and matching-adjustedsecond-line CELESTIAL populations compared with the

RESORCE population. aUnanchored analysis. CI confi-dence interval, OR odds ratio, TEAE treatment-emergentadverse event

2690 Adv Ther (2020) 37:2678–2695

Clinical Interpretation

Despite these limitations, when evidence fromhead-to-head trials is not available, a MAICoffers insight into the plausible comparativeresults for different therapies in a commonclinical context, which can be valuable forhealthcare professionals and health technologydecision-makers. Such insights are particularlyimportant when selecting the optimum treat-ment approach for patients with advanced HCCfor whom the prognosis has traditionally beenpoor, with few therapeutic options available.First-line treatment in advanced HCC fre-quently fails after a period of time owing toadaptive or intrinsic resistance, diseaseprogression or significant toxicity [17]: thus,there is a growing need for second- and later-line treatment options and for insights that willhelp to guide clinicians when selecting theoptimum treatment sequence for their patients.

From a clinical perspective, the results of thisanalysis reinforce and further those of a recentnetwork meta-analysis of randomizedcontrolled trials of second-line agents for use inHCC after prior sorafenib [38]. The networkmeta-analysis included 13 trials and 11 differentsecond-line therapies for advanced HCC andgenerated anchored efficacy and safetyestimates using everolimus as the commoncomparator with OS as the primary endpoint.Of the 11 treatments included in the analysis,only cabozantinib and regorafenib significantlyprolonged OS [38]. Overall, the authors con-cluded that second-line cabozantinib and rego-rafenib offered the best combination of efficacyand safety for patients with advanced HCC fol-lowing prior sorafenib therapy. They furthernoted that the role of cabozantinib may be ofparticular importance in patients who wereintolerant to sorafenib, because sorafenib-in-tolerant patients were excluded from theRESORCE trial of regorafenib [38]. In clinicalpractice, a range of factors must inform deci-sion-making in optimum treatment sequenc-ing. As well as the anticipated effectiveness andtolerability, treatment selection is influenced bypatient characteristics (e.g., comorbidities) andtheir preferences, route of administration, dos-ing regimen and implications for patient

quality of life. Also worthy of consideration area patient’s response to and tolerance of priortreatment as well as the molecular targets ofavailable therapeutic options and whether theycomplement those of prior treatment andaddress possible mechanisms of resistance (e.g.,AXL, MER) [20, 21].

CONCLUSION

MAIC analyses permit indirect comparisonsbetween clinical trials with heterogeneouspopulations but with common treatment out-comes. In this MAIC analysis, which used datafrom the CELESTIAL and RESORCE trials,unanchored survival estimates suggestcabozantinib may be associated with similar OSand prolonged PFS compared with regorafenibin patients with advanced progressive HCC whowere receiving second-line treatment followingprogression after prior sorafenib; regorafenibwas associated with lower rates of diarrhea.Although this MAIC provides a useful indica-tion of the comparative efficacy and safety ofthese two second-line agents, it is not areplacement for a head-to-head comparativetrial.

ACKNOWLEDGEMENTS

The authors thank all patients involved in thesestudies as well as their caregivers, care team,investigators and research staff in participatinginstitutions.

Funding. This study, manuscript develop-ment and the journal’s Rapid Service andOpen Access fees were funded by Ipsen. TheCELESTIAL study was sponsored by Exelixis.The sponsor was involved in the design of thestudy, analysis and interpretation as well asreview of the manuscript.

Medical Writing, Editorial and OtherAssistance. The authors thank Antonio RemiroAzocar (PhD) and Professor Gianluca Baio ofUniversity College London, London, UK, fortheir contribution to the study analysis and

Adv Ther (2020) 37:2678–2695 2691

Alison Chisholm (MPH) and Tamzin Gristwood(PhD) of Oxford PharmaGenesis, Oxford, UK,who provided medical writing and editorialsupport, which was sponsored by Ipsen inaccordance with Good Publication Practiceguidelines.

Authorship. All named authors meet theInternational Committee of Medical JournalEditors (ICMJE) criteria for authorship for thisarticle, take responsibility for the integrity ofthe work as a whole and have given theirapproval for this version to be published.

Disclosures. GK Abou-Alfa: Research grantsfrom ActaBiologica, Agios, Array, AstraZeneca,Bayer, Beigene, Bristol-Myers Squibb, CasiPharmaceuticals, Celgene, Exelixis, Genentech,Halozyme, Incyte, Mabvax, Polaris Puma, QEDTherapeutics and Roche; consultancy fees fromAgios, AstraZeneca, Autem, Bayer, Beigene,Berry Genomics, Bioline, Bristol-Myers Squibb,Celgene, CytomX, Debio, Eisai, Exelixis, Fla-tiron, Genoscience, Incyte, Ipsen, Jansen, LAM,Lilly, Loxo Oncology, Merck, MinapharmPharmaceuticals, Pfizer, QED Therapeutics,RedHill Biopharma, Silenseed, Sillajen, Sobi,Targovax, Therabionics, Twoxar and Yiviva. J-FBlanc: Consultancy fees from: Bayer, Bristol-Myers Squibb, Esai Co. Ltd, Lilly, Ipsen andOnxeo. A-L Cheng: Consultancy fees fromAstraZeneca, Bayer Schering Pharma, BayerYakuhin, Bristol-Myers Squibb, CSR PharmaGroup Inc., Eisai, Eli Lilly, Genentech/Roche,Merck Sharp & Dohme, Novartis and OnoPharmaceutical; travel grants from Bayer Yaku-hin, Eisai, Genentech/Roche and Roche (Tai-wan). B Daniele: Personal fees and non-financialsupport from Bayer, Ipsen and Sanofi; personalfees from AstraZeneca, Eli Lilly, Esai Co. Ltd,Incyte, Merck Sharp & Dohme and Roche.I Guerra: employee of IQVIA, which was con-tracted by Ipsen to conduct the analysis. RKKelley: Research grants (to institution) for clin-ical trials from Adaptimmune, Agios, AstraZe-neca, Bayer, Bristol-Myers Squibb, Eli Lilly, EMDSerono, Exelixis, Merck, Novartis, PartnerTherapeutics, QED Therapeutics, and TaihoPharmaceutical Group; consultancy fees (toinstitution) from Agios, AstraZeneca, and

Bristol-Myers Squibb; consultancy fees (to self)from Genentech/Roche and Gilead; travel sup-port (to self) from Ipsen. Thomas Yau: Consul-tancy fees from Bayer, Bristol-Myers Squibb,Eisai and Merck Sharp & Dohme. F Marteau, PMollon and V Valcheva: employees of Ipsen.

Compliance with Ethics Guidelines. Theresults presented in this manuscript are basedon previously published studies. All proceduresperformed in those studies involving humanparticipants were in accordance with the ethicalstandards of the local Institutional ReviewBoards for each site and with the 1964 HelsinkiDeclaration and its later amendments or com-parable ethical standards. Informed consent wasobtained from all individual participants inclu-ded in the CELESTIAL (ClinicalTrials.gov iden-tifier NCT01908426) and RESORCE(ClinicalTrials.gov identifier NCT01774344)trials.

Data Availability. Where patient data canbe anonymized, Ipsen will share all individualparticipant data that underlie the results repor-ted in this article with qualified researchers whoprovide a valid research question. Study docu-ments, such as the study protocol and clinicalstudy report, are not always available. Proposalsshould be submitted to DataSharing@Ipsen.comand will be assessed by a scientific review board.Data are available beginning 6 months andending 5 years after publication; after this time,only raw data may be available.

Open Access. This article is licensed undera Creative Commons Attribution-NonCommer-cial 4.0 International License, which permitsany non-commercial use, sharing, adaptation,distribution and reproduction in any mediumor format, as long as you give appropriate creditto the original author(s) and the source, providea link to the Creative Commons licence, andindicate if changes were made. The images orother third party material in this article areincluded in the article’s Creative Commonslicence, unless indicated otherwise in a creditline to the material. If material is not includedin the article’s Creative Commons licence andyour intended use is not permitted by statutory

2692 Adv Ther (2020) 37:2678–2695

regulation or exceeds the permitted use, youwill need to obtain permission directly from thecopyright holder. To view a copy of this licence,visit http://creativecommons.org/licenses/by-nc/4.0/.

REFERENCES

1. European Association for the Study of the LiverClinical Practice Guidelines: management of hepa-tocellular carcinoma. J Hepatol. 2018;69(1):182–236. https://doi.org/10.1016/j.jhep.2018.03.019.

2. Llovet JM, Ricci S, Mazzaferro V, Hilgard P, Gane E,Blanc J-F, et al. Sorafenib in advanced hepatocellu-lar carcinoma. New Engl J Med. 2008;359(4):378–90. https://doi.org/10.1056/NEJMoa0708857.

3. Bayer Schering Pharma. Nexavar� first FDA-ap-proved drug therapy for liver cancer. 2007. https://www.investor.bayer.de/en/nc/news/archive/investor-news-2007/investor-news-2007/nexavarR-first-fda-approved-drug-therapy-for-liver-cancer/. Accessed17 April 2020.

4. Kudo M. Targeted therapy for liver cancer: updatedreview in 2012. Curr Cancer Drug Targets.2012;12(9):1062–72.

5. Kudo M, Finn RS, Qin S, Han K-H, Ikeda K, PiscagliaF, et al. Lenvatinib versus sorafenib in first-linetreatment of patients with unresectable hepatocel-lular carcinoma: a randomised phase 3 non-inferi-ority trial. Lancet. 2018;391(10126):1163–73.https://doi.org/10.1016/S0140-6736(18)30207-1.

6. Merck & Co. Inc. Eisai and Merck announce FDAapproval of Lenvima� (lenvatinib) capsules for first-line treatment of unresectable hepatocellular carci-noma (HCC). 2018. https://investors.merck.com/news/press-release-details/2018/Eisai-And-Merck-Announce-FDA-Approval-Of-LENVIMA-lenvatinib-Capsules-For-First-line-Treatment-Of-Unresectable-Hepatocellular-Carcinoma-HCC/default.aspx.Accessed 17 April 2020.

7. Bayer. Bayer receives FDA approval for stivarga�(regorafenib) for the second-line systemic treat-ment of liver cancer. 2017. https://www.bayer.us/en/newsroom/press-releases/article/?id=123100.Accessed 17 April 2020.

8. Bruix J, Qin S, Merle P, Granito A, Huang Y-H,Bodoky G, et al. Regorafenib for patients withhepatocellular carcinoma who progressed on sor-afenib treatment (RESORCE): a randomised,

double-blind, placebo-controlled, phase 3 trial.Lancet. 2017;389(10064):56–66. https://doi.org/10.1016/S0140-6736(16)32453-9.

9. Ipsen. Ipsen receives positive CHMP opinion forCabometyx� (cabozantinib) for the second-linetreatment of patients with hepatocellular carcinoma(HCC). 2018. https://www.ipsen.com/press-releases/ipsen-receives-positive-chmp-opinion-for-cabometyx-cabozantinib-for-the-second-line-treatment-of-patients-with-hepatocellular-carcinoma-hcc/.Accessed 17 April 2020.

10. Abou-Alfa GK, Meyer T, Cheng A-L, El-Khoueiry AB,Rimassa L, Ryoo B-Y, et al. Cabozantinib in patientswith advanced and progressing hepatocellular car-cinoma. New Engl J Med. 2018;379(1):54–63.https://doi.org/10.1056/NEJMoa1717002.

11. Eli Lilly and Company. Lilly’s Cyramza� (ramu-cirumab) becomes first FDA-approved biomarker-driven therapy in patients with hepatocellular car-cinoma. 2019. https://investor.lilly.com/news-releases/news-release-details/lillys-cyramzar-ramucirumab-becomes-first-fda-approved-biomarker.Accessed 17 April 2020.

12. Zhu AX, Park JO, Ryoo B-Y, Yen C-J, Poon R, Pas-torelli D, et al. Ramucirumab versus placebo assecond-line treatment in patients with advancedhepatocellular carcinoma following first-line ther-apy with sorafenib (REACH): a randomised, double-blind, multicentre, phase 3 trial. Lancet Oncol.2015;16(7):859–70. https://doi.org/10.1016/S1470-2045(15)00050-9.

13. Bristol-Myers Squibb. Prescribing Informaton:Opdivo (nivolumab) injection, for intravenous use.2020. https://www.accessdata.fda.gov/drugsatfda_docs/label/2020/125554s078lbl.pdf. Accessed 17April 2020.

14. Bristol-Myers Squibb. Prescribing Information: Yer-voy� (ipilimumab) injection, for intravenous use.2020 https://www.accessdata.fda.gov/drugsatfda_docs/label/2020/125377s108lbl.pdf. Accessed 17April 2020.

15. Merck & Co. Inc. FDA Approves Merck’s KEY-TRUDA� (pembrolizumab) for the Treatment ofPatients with Hepatocellular Carcinoma (HCC) WhoHave Been Previously Treated with Sorafenib. https://investors.merck.com/news/press-release-details/2018/FDA-Approves-Mercks-KEYTRUDA-pembrolizumab-for-the-Treatment-of-Patients-with-Hepatocellular-Carcinoma-HCC-Who-Have-Been-Previously-Treated-with-Sorafenib/default.aspx. Accessed 17 April 2020.

16. Finn RS, Ryoo BY, Merle P, Kudo M, Bouattour M,Lim HY, et al. Pembrolizumab as second-line ther-apy in patients with advanced hepatocellular car-cinoma in KEYNOTE-240: a randomized, double-

Adv Ther (2020) 37:2678–2695 2693

http://creativecommons.org/licenses/by-nc/4.0/http://creativecommons.org/licenses/by-nc/4.0/https://doi.org/10.1016/j.jhep.2018.03.019https://doi.org/10.1016/j.jhep.2018.03.019https://doi.org/10.1056/NEJMoa0708857https://www.investor.bayer.de/en/nc/news/archive/investor-news-2007/investor-news-2007/nexavarR-first-fda-approved-drug-therapy-for-liver-cancer/https://www.investor.bayer.de/en/nc/news/archive/investor-news-2007/investor-news-2007/nexavarR-first-fda-approved-drug-therapy-for-liver-cancer/https://www.investor.bayer.de/en/nc/news/archive/investor-news-2007/investor-news-2007/nexavarR-first-fda-approved-drug-therapy-for-liver-cancer/https://www.investor.bayer.de/en/nc/news/archive/investor-news-2007/investor-news-2007/nexavarR-first-fda-approved-drug-therapy-for-liver-cancer/https://doi.org/10.1016/S0140-6736(18)30207-1https://investors.merck.com/news/press-release-details/2018/Eisai-And-Merck-Announce-FDA-Approval-Of-LENVIMA-lenvatinib-Capsules-For-First-line-Treatment-Of-Unresectable-Hepatocellular-Carcinoma-HCC/default.aspxhttps://investors.merck.com/news/press-release-details/2018/Eisai-And-Merck-Announce-FDA-Approval-Of-LENVIMA-lenvatinib-Capsules-For-First-line-Treatment-Of-Unresectable-Hepatocellular-Carcinoma-HCC/default.aspxhttps://investors.merck.com/news/press-release-details/2018/Eisai-And-Merck-Announce-FDA-Approval-Of-LENVIMA-lenvatinib-Capsules-For-First-line-Treatment-Of-Unresectable-Hepatocellular-Carcinoma-HCC/default.aspxhttps://investors.merck.com/news/press-release-details/2018/Eisai-And-Merck-Announce-FDA-Approval-Of-LENVIMA-lenvatinib-Capsules-For-First-line-Treatment-Of-Unresectable-Hepatocellular-Carcinoma-HCC/default.aspxhttps://investors.merck.com/news/press-release-details/2018/Eisai-And-Merck-Announce-FDA-Approval-Of-LENVIMA-lenvatinib-Capsules-For-First-line-Treatment-Of-Unresectable-Hepatocellular-Carcinoma-HCC/default.aspxhttps://www.bayer.us/en/newsroom/press-releases/article/?id=123100https://www.bayer.us/en/newsroom/press-releases/article/?id=123100https://doi.org/10.1016/S0140-6736(16)32453-9https://doi.org/10.1016/S0140-6736(16)32453-9https://www.ipsen.com/press-releases/ipsen-receives-positive-chmp-opinion-for-cabometyx-cabozantinib-for-the-second-line-treatment-of-patients-with-hepatocellular-carcinoma-hcc/https://www.ipsen.com/press-releases/ipsen-receives-positive-chmp-opinion-for-cabometyx-cabozantinib-for-the-second-line-treatment-of-patients-with-hepatocellular-carcinoma-hcc/https://www.ipsen.com/press-releases/ipsen-receives-positive-chmp-opinion-for-cabometyx-cabozantinib-for-the-second-line-treatment-of-patients-with-hepatocellular-carcinoma-hcc/https://www.ipsen.com/press-releases/ipsen-receives-positive-chmp-opinion-for-cabometyx-cabozantinib-for-the-second-line-treatment-of-patients-with-hepatocellular-carcinoma-hcc/https://doi.org/10.1056/NEJMoa1717002https://investor.lilly.com/news-releases/news-release-details/lillys-cyramzar-ramucirumab-becomes-first-fda-approved-biomarkerhttps://investor.lilly.com/news-releases/news-release-details/lillys-cyramzar-ramucirumab-becomes-first-fda-approved-biomarkerhttps://investor.lilly.com/news-releases/news-release-details/lillys-cyramzar-ramucirumab-becomes-first-fda-approved-biomarkerhttps://doi.org/10.1016/S1470-2045(15)00050-9https://doi.org/10.1016/S1470-2045(15)00050-9https://www.accessdata.fda.gov/drugsatfda_docs/label/2020/125554s078lbl.pdfhttps://www.accessdata.fda.gov/drugsatfda_docs/label/2020/125554s078lbl.pdfhttps://www.accessdata.fda.gov/drugsatfda_docs/label/2020/125377s108lbl.pdfhttps://www.accessdata.fda.gov/drugsatfda_docs/label/2020/125377s108lbl.pdfhttps://investors.merck.com/news/press-release-details/2018/FDA-Approves-Mercks-KEYTRUDA-pembrolizumab-for-the-Treatment-of-Patients-with-Hepatocellular-Carcinoma-HCC-Who-Have-Been-Previously-Treated-with-Sorafenib/default.aspxhttps://investors.merck.com/news/press-release-details/2018/FDA-Approves-Mercks-KEYTRUDA-pembrolizumab-for-the-Treatment-of-Patients-with-Hepatocellular-Carcinoma-HCC-Who-Have-Been-Previously-Treated-with-Sorafenib/default.aspxhttps://investors.merck.com/news/press-release-details/2018/FDA-Approves-Mercks-KEYTRUDA-pembrolizumab-for-the-Treatment-of-Patients-with-Hepatocellular-Carcinoma-HCC-Who-Have-Been-Previously-Treated-with-Sorafenib/default.aspxhttps://investors.merck.com/news/press-release-details/2018/FDA-Approves-Mercks-KEYTRUDA-pembrolizumab-for-the-Treatment-of-Patients-with-Hepatocellular-Carcinoma-HCC-Who-Have-Been-Previously-Treated-with-Sorafenib/default.aspxhttps://investors.merck.com/news/press-release-details/2018/FDA-Approves-Mercks-KEYTRUDA-pembrolizumab-for-the-Treatment-of-Patients-with-Hepatocellular-Carcinoma-HCC-Who-Have-Been-Previously-Treated-with-Sorafenib/default.aspxhttps://investors.merck.com/news/press-release-details/2018/FDA-Approves-Mercks-KEYTRUDA-pembrolizumab-for-the-Treatment-of-Patients-with-Hepatocellular-Carcinoma-HCC-Who-Have-Been-Previously-Treated-with-Sorafenib/default.aspx

blind. Phase III Trial J Clin Oncol. 2020;38(3):193–202. https://doi.org/10.1200/jco.19.01307.

17. Ray EM, Sanoff HK. Optimal therapy for patientswith hepatocellular carcinoma and resistance orintolerance to sorafenib: challenges and solutions.J Hepatocell Carcinoma. 2017;4:131–8. https://doi.org/10.2147/JHC.S124366.

18. American Cancer Society. Liver cancer survivalrates. 2020. https://www.cancer.org/cancer/liver-cancer/detection-diagnosis-staging/survival-rates.html. Accessed 17 April 2020.

19. Bayer Plc. Summary of product characteristics: Sti-varga 40 mg film-coated tablets. 2018. https://www.medicines.org.uk/emc/product/1263/smpc. Acces-sed 17 April 2020.

20. Ipsen Ltd. Summary of Product Characteristics:Cabometyx 20 mg Film-coated Tablets. 2019.https://www.medicines.org.uk/emc/product/4331/smpc. Accessed 17 April 2020.

21. Mukherjee SK, Wilhelm A, Antoniades CG. TAMreceptor tyrosine kinase function and theimmunopathology of liver disease. Am J PhysiolGastrointest Liver Physiol. 2016;310(11):G899–G905. https://doi.org/10.1152/ajpgi.00382.2015.

22. Vogel A, Cervantes A, Chau I, Daniele B, Llovet J,Meyer T, et al. Hepatocellular carcinoma: ESMOclinical practice guidelines for diagnosis, treatmentand follow-up. Ann Oncol. 2018;29(Suppl 4):iv238–iv55. https://doi.org/10.1093/annonc/mdy308.

23. National Comprehensive Cancer Network. NCCNclinical practice guidelines in onclogy (NCCNGuidelines�): hepatobiliary cancers. Version 1.2018 - February 14, 2018. https://www.nccn.org/professionals/physician_gls/pdf/hepatobiliary.pdf.Accessed 17 April 2020.

24. Thom H, Jugl SM, Palaka E, Jawla S. Matchingadjusted indirect comparisons to assess compara-tive effectiveness of therapies: usage in scientificliterature and health technology appraisals. ValueHealth. 2016;19(3):A100–A101101. https://doi.org/10.1016/j.jval.2016.03.1723.

25. Phillippo DM AA, Dias S, Palmer S, Abrams KR,Welton NJ. NICE DSU technical support document18: methods for population-adjusted indirect com-parisons in submissions to NICE: 2016. https://www.nicedsu.org.uk/wp-content/uploads/2017/05/Population-adjustment-TSD-FINAL.pdf. Accessed17 April 2020.

26. Salanti G, Del Giovane C, Chaimani A, CaldwellDM, Higgins JPT. Evaluating the quality of evidence

from a network meta-analysis. PLoS One. 2014;9(7):e99682–e. https://doi.org/10.1371/journal.pone.0099682.

27. Chowdhury S, Oudard S, Uemura H, Joniau S, PilonD, Lefebvre P, et al. Matching-adjusted indirectcomparison of health-related quality of life andadverse events of apalutamide versus enzalutamidein non-metastatic castration-resistant prostate can-cer. Adv Ther. 2020;37(1):512–26. https://doi.org/10.1007/s12325-019-01157-4.

28. Tremblay G, Chandiwana D, Dolph M, Hearnden J,Forsythe A, Monaco M. Matching-adjusted indirecttreatment comparison of ribociclib and palbociclibin HR?, HER2- advanced breast cancer. CancerManag Res. 2018;10:1319–27. https://doi.org/10.2147/CMAR.S163478.

29. Odom D, Mladsi D, Purser M, Kaye JA, Palaka E,Charter A, et al. A matching-adjusted indirectcomparison of sonidegib and vismodegib inadvanced basal cell carcinoma. J Skin Cancer. 2017.https://doi.org/10.1155/2017/6121760.

30. Venkatachalam M, Stenehjem D, Parikh ND, SinghP, Marett B, Sill B, et al. Cost of treatment-relatedadverse events (TRAES) in second-line (2l) advancedhepatocellular carcinoma (AHCC): match adjustedindirect comparison (MAIC) of nivolumab andregorafenib. Value Health. 2017;20(9):A502–A503503. https://doi.org/10.1016/j.jval.2017.08.590.

31. Roskell N, Gregory J, Wisniewski T, Thompson GJ,De la Cruz C. Indirect treatment comparision ofnivolumab versu regorafenib, cabozantinib and bestsupportive care after treatent with sorafenib forhepatocellular carcinoma (PCN17). Value Health.2018;21:S17–S1818. https://doi.org/10.1016/j.jval.2018.09.099.

32. Parikh N, Marshall A, Huff KD, Savidge R, Betts KA,Song J, et al. Comparative efficacy of second-linetreatments for advanced hepatocellular carcinoma:a network meta-analysis. J Clin Oncol.2020;38(4_suppl):545. https://doi.org/10.1200/JCO.2020.38.4_suppl.545.

33. Latimer N. NICE DSU technical support document14: survival analysis for economic evaluationsalongside clinical trials-extrapolation with patient-level data: 2011. https://nicedsu.org.uk/wp-content/uploads/2016/03/NICE-DSU-TSD-Survival-analysis.updated-March-2013.v2.pdf. Accessed 17April 2020.

34. Cole SR, Hernán MA. Adjusted survival curves withinverse probability weights. Comput Methods Pro-grams Biomed. 2004;75(1):45–9. https://doi.org/10.1016/j.cmpb.2003.10.004.

2694 Adv Ther (2020) 37:2678–2695

https://doi.org/10.1200/jco.19.01307https://doi.org/10.2147/JHC.S124366https://doi.org/10.2147/JHC.S124366https://www.cancer.org/cancer/liver-cancer/detection-diagnosis-staging/survival-rates.htmlhttps://www.cancer.org/cancer/liver-cancer/detection-diagnosis-staging/survival-rates.htmlhttps://www.cancer.org/cancer/liver-cancer/detection-diagnosis-staging/survival-rates.htmlhttps://www.medicines.org.uk/emc/product/1263/smpchttps://www.medicines.org.uk/emc/product/1263/smpchttps://www.medicines.org.uk/emc/product/4331/smpchttps://www.medicines.org.uk/emc/product/4331/smpchttps://doi.org/10.1152/ajpgi.00382.2015https://doi.org/10.1152/ajpgi.00382.2015https://doi.org/10.1093/annonc/mdy308https://doi.org/10.1093/annonc/mdy308https://www.nccn.org/professionals/physician_gls/pdf/hepatobiliary.pdfhttps://www.nccn.org/professionals/physician_gls/pdf/hepatobiliary.pdfhttps://doi.org/10.1016/j.jval.2016.03.1723https://doi.org/10.1016/j.jval.2016.03.1723http://www.nicedsu.org.uk/wp-content/uploads/2017/05/Population-adjustment-TSD-FINAL.pdfhttp://www.nicedsu.org.uk/wp-content/uploads/2017/05/Population-adjustment-TSD-FINAL.pdfhttp://www.nicedsu.org.uk/wp-content/uploads/2017/05/Population-adjustment-TSD-FINAL.pdfhttps://doi.org/10.1371/journal.pone.0099682https://doi.org/10.1371/journal.pone.0099682https://doi.org/10.1007/s12325-019-01157-4https://doi.org/10.1007/s12325-019-01157-4https://doi.org/10.2147/CMAR.S163478https://doi.org/10.2147/CMAR.S163478https://doi.org/10.1155/2017/6121760https://doi.org/10.1016/j.jval.2017.08.590https://doi.org/10.1016/j.jval.2017.08.590https://doi.org/10.1016/j.jval.2018.09.099https://doi.org/10.1016/j.jval.2018.09.099https://doi.org/10.1200/JCO.2020.38.4_suppl.545https://doi.org/10.1200/JCO.2020.38.4_suppl.545http://nicedsu.org.uk/wp-content/uploads/2016/03/NICE-DSU-TSD-Survival-analysis.updated-March-2013.v2.pdfhttp://nicedsu.org.uk/wp-content/uploads/2016/03/NICE-DSU-TSD-Survival-analysis.updated-March-2013.v2.pdfhttp://nicedsu.org.uk/wp-content/uploads/2016/03/NICE-DSU-TSD-Survival-analysis.updated-March-2013.v2.pdfhttps://doi.org/10.1016/j.cmpb.2003.10.004https://doi.org/10.1016/j.cmpb.2003.10.004

35. Link CL. Confidence intervals for the survivalfunction using Cox’s proportional-hazard modelwith covariates. Biometrics. 1984:601–9. https://doi.org/10.2307/2530904; https://www.jstor.org/stable/2530904.

36. Woodruff RS. Confidence intervals for medians andother position measures. J Am Stat Assoc.1952;47(260):635–46. https://doi.org/10.1080/01621459.1952.10483443.

37. Xue Y, Schifano ED. Diagnostics for the Cox model.Commun Stat Appl Methods. 2017;24(6):583–604.https://doi.org/10.29220/CSAM.2017.24.6.583.

38. Bakouny Z, Assi T, El Rassy E, Nasr F. Second-linetreatments of advanced hepatocellular carcinoma:systematic review and network meta-analysis ofrandomized controlled trials. J Clin Gastroenterol.2019;53(4):251–61. https://doi.org/10.1097/MCG.0000000000001160.

39. Schmidt E, Lister J, Neumann M, Wiecek W, Fu S,Vataire AL, et al. Cabozantinib versus standard-of-care comparators in the treatment of advanced/metastatic renal cell carcinoma in treatment-naivepatients: a systematic review and network meta-analysis. Target Oncol. 2018;13(2):205–16. https://doi.org/10.1007/s11523-018-0559-0.

Adv Ther (2020) 37:2678–2695 2695

https://doi.org/10.2307/2530904https://doi.org/10.2307/2530904https://www.jstor.org/stable/2530904https://www.jstor.org/stable/2530904https://doi.org/10.1080/01621459.1952.10483443https://doi.org/10.1080/01621459.1952.10483443https://doi.org/10.29220/CSAM.2017.24.6.583https://doi.org/10.1097/MCG.0000000000001160https://doi.org/10.1097/MCG.0000000000001160https://doi.org/10.1007/s11523-018-0559-0https://doi.org/10.1007/s11523-018-0559-0

Comparative Efficacy of Cabozantinib and Regorafenib for Advanced Hepatocellular CarcinomaAbstractBackgroundObjectivesMethodsResultsConclusions

Plain Language SummaryIntroductionMethodsData Source Identification And EligibilityMatching-Adjusted Indirect Comparison: Cabozantinib and RegorafenibPopulation MatchingOutcome Evaluation: Survival AnalysisOutcome Evaluation: Safety

AnalysesCompliance with Ethics Guidelines

ResultsPatient CharacteristicsSurvival OutcomesKM-Derived EstimatesParametric Modeling Estimates

Safety Outcomes

DiscussionOutcome InterpretationLimitationsClinical Interpretation

ConclusionAcknowledgementsReferences