Trends of Clinical Outcomes of Patients with Advanced Hepatocellular Carcinoma Treated with First-Line Sorafenib in Randomized Controlled Trials

Hepatocellular carcinoma (HCC) is a common cancer globally and is associated with considerable morbidity and mortality [1]. Although early HCC can be managed with potentially curative locoregional therapies such as resection or ablation or with liver transplantation, those who present with advanced disease have limited treatment options and frequently have a poor prognosis [2].

Sorafenib was the first systemic therapy approved by the United States Food and Drug Administration for the treatment of advanced HCC on the basis of clinical trials demonstrating a statistically significant survival benefit compared to placebo [3, 4]. However, sorafenib only has modest antitumor activity and its use is limited by significant toxicity, with side effects driving discontinuation of therapy in nearly 40% of patients [5]. Approximately a decade after the original approval of sorafenib, the REFLECT trial demonstrated that lenvatinib was noninferior to sorafenib with regard to overall survival (OS) and superior to sorafenib with certain secondary outcome measures; lenvatinib became an acceptable initial therapy for advanced HCC [6, 7]. Recently, a trial of the combination of atezolizumab and bevacizumab demonstrated significantly prolonged OS compared to sorafenib in patients with advanced HCC and Child-Pugh (CP) A cirrhosis, and also demonstrated improvements in multiple secondary endpoints [8]. For patients that qualify for this regimen, combination atezolizumab and bevacizumab has supplanted sorafenib as a preferred frontline option in many parts of the world [8]. Despite this, sorafenib remains a standard control arm for most ongoing global clinical trials in frontline advanced HCC, in part because many of these studies were initiated prior to the adoption of combination atezolizumab-bevacizumab as standard frontline therapy.

The use of a single standard systemic therapy arm in multiple global late-phase clinical trials spanning over a decade provides a unique opportunity to develop insights into clinical practices and clinical trials conduct of advanced HCC and potential non-drug factors that impact outcomes. Further, since the initial approval of sorafenib as standard therapy for advanced HCC, multidisciplinary care for all patients with HCC has become more widely accepted and recognized to improve patient outcomes [9, 10]. Additionally, earlier diagnosis, institution of supportive care, recognition and mitigation of side effects of sorafenib, earlier initiation of systemic therapy, and treatment of hepatitis C may also contribute to improved outcomes in patients with advanced HCC [9-13]. In this study, we explored the sorafenib comparator arms in clinical trials of previously untreated advanced HCC and hypothesized that the OS of patients receiving sorafenib has increased in published comparative trials of systemic treatment-naive HCC.

We searched PubMed to identify relevant randomized clinical trials of sorafenib in patients with advanced HCC naive to systemic therapy. Trials were included if they included patients were naive to systemic therapy, presented outcomes of interest, and had a comparison arm of systemic therapy (active therapy or placebo-controlled). Studies that compared sorafenib to a procedure (e.g., hepatic artery infusion or radiofrequency ablation) were not included. Studies that appeared to meet criteria by abstract review were examined more closely for inclusion. Bibliographies of included studies were then hand-searched for other studies that could potentially meet inclusion criteria. After development of a preliminary list of included studies, we approached content experts (authors T.B.K. and M.Y.) for further identification of studies.

We extracted demographic data and trial-level outcome data from the resulting publications, meeting abstracts, or publicly available protocol information. Outcomes of interest included median OS, median progression-free survival (PFS), objective response rates (ORR), and median duration of therapy of included patients receiving sorafenib. ORRs were separated by Response Evaluation Criteria in Solid Tumors (RECIST) 1.0, 1.1, or modified RECIST (mRECIST). Publication bias was assessed by visual inspection of a funnel plot of sample size versus median OS (the primary outcome of interest). Extracted data were assessed for normality using skewness and kurtosis testing and joint χ² testing with a p value of less than 0.05 indicating rejection of the null hypothesis of normal distribution. Comparisons of demographics and outcomes were made between phase II and phase III trials using Wilcoxon Rank Sum testing for non-normally distributed factors or 2-sample t- tests if normally distributed.

We performed weighted linear regression analysis to identify temporal trends while accounting for differences in sample sizes. Temporal data from studies were analyzed from the date of first patient enrollment as this was felt to be most representative of contemporary supportive care at the time the respective trials were performed. Data were analyzed in Stata version 17.0. Outliers, when identified, were formally tested with Cook’s distance. Analyses were also stratified by trial phase (II vs. III). Significance was set at p ≤ 0.05; no adjustments were made for multiple comparisons due to the exploratory nature of this analysis.

This study did not use any identifiable patient data and only utilized previously published publicly available data. No specific informed consent was sought. This work was supported by the grants from the National Institutes of Health (T32CA009679 to T.J.B., NCI SPORE P50 CA062924, NIH P30 CA006973, and NIH UM1CA186691 all to M.Y.).

A total of 100 studies were included in the initial database screening. Ultimately, 14 studies of systemic therapies in advanced HCC using sorafenib versus placebo or other systemic therapy were identified (6 phase II and 8 phase III); 1 additional phase III study that has only been published in abstract form was identified from discussion with experts (Checkmate-459) [3, 4, 6, 8, 14-24]. During data analysis and manuscript preparation, results from 1 additional phase II clinical trial of sorafenib in HCC were published and subsequently included in the analysis [25]. In total, 16 trials were included in at least 1 analysis (Table 1).

A total of 4,086 patients were enrolled from 2005 to 2019 to receive sorafenib in comparison studies of treatment-naive HCC. The median age of patients included was 60.4 years (95% confidence interval 57.5–62.3), 82.4% of participants were male, 55.4% had an Eastern Cooperative Oncology Group (ECOG) performance status (PS) of 0, and 43.2% had an ECOG PS 1. Most patients were staged with Barcelona Clinic Liver Cancer (BCLC) stage C HCC and most studies only included patients with CP A cirrhosis.

Phase III trials were larger (p = 0.003) and contained a higher proportion of patients with extrahepatic spread (p = 0.028); however, there were no other significant differences in the proportions of extracted demographics (ECOG0, ECOG 1, CP A cirrhosis, CP B cirrhosis, hepatitis B, hepatitis C, prior locoregional therapies, or vascular invasion) between trial phases. No significant differences were noted in median OS, median PFS by RECIST 1.1, ORR by RECIST 1.1, or median duration of therapy between phase II and phase III studies (see online suppl. Table 1; for all online suppl. material, see www.karger.com/doi/10.1159/000521625).

No linear trends were identified by sample-weighted linear regression between the date of enrollment and proportion of patients with CP A cirrhosis, BCLC B HCC, median age, proportion of ECOG 1 PS, proportion of patients with HBV, or proportion of patients with HCV. There was a trend toward a higher proportion of ECOG0 patients enrolled in more contemporary studies; however, this was not statistically significant (p = 0.084, R² = 0.17). There was also a positive trend of date of enrollment and proportion of patients who have previously received locoregional therapies; however, this also did not reach statistical significance (p = 0.08, R² = 0.368). In phase III trials, there was a nonsignificant trend toward an increased proportion of patients with CP A cirrhosis (p = 0.08, R² 0.098); however, no other demographic trend approached significance by weighted linear regression.

The sample-weighted median OS of all patients receiving sorafenib was 10.1 months (95% CI 9.0–11.6 months, 16 studies), and the sample-weighted median PFS by RECIST 1.1 was 3.8 months (95% CI 3.5–4.9 months, 13 studies) and 4.1 months (95% CI 3.1–4.9 months, 3 studies) by mRECIST. The median of reported median OS weighted by participants was 10.0 months for phase II studies (95% CI 7.6–12.1 months) and 10.2 months (95% CI 8.7–12.5 months) for phase III studies. No linear trend was noted for sample size by year of enrollment (p = 0.89, R² = 0.002).

All 16 included studies comprising 4,086 patients contributed data on median OS. A significant positive linear trend was noted between date of first enrollment and OS (p = 0.048, R² = 0.38) indicating improvement in median OS of 4.5 months over time favoring the more contemporary studies after accounting for differences in sample sizes. This amounted to a median increase of 0.34 months improvement in OS each year since the first patient was enrolled, or 10.5 days improvement in OS per year (Table 1; Fig. 1). When analyzed by trial phase, significant linear trends were again noted. Phase III studies (9 studies, 3,462 patients) had a positive linear trend (p = 0.02) with a slope of 0.47 months per year enrolled (R² = 0.64). No significant trend of OS and date of enrollment was identified when this analysis was limited to phase II trials (p = 0.92, R² = 0.002, 7 studies, 624 patients).

Thirteen studies (3,374 patients) reported PFS by RECIST 1.1. No linear relationship was noted between median PFS by RECIST 1.1 with sorafenib by year of enrollment weighted by enrolled patients (Fig. 2, p = 0.69, R² 0.021). When this analysis was stratified by phase, again no significant linear relationship with year of enrollment was identified. However, visual inspection of the scatterplot of median PFS by RECIST 1.1 and date of enrollment in phase III trials revealed the median PFS reported by SHARP (5.5 months with first date of enrollment of March 10, 2005) to likely be an outlier, which was confirmed by Cook’s distance. An exploratory analysis that dropped SHARP from the weighted linear regression model resulted in improved R² from 0.0135 to 0.3153; however, the resulting positive trend was not statistically significant (p = 0.11). There were too few studies reporting PFS by mRECIST to perform an analogous analysis.

Fourteen studies of 3,457 patients reported on ORR by RECIST 1.1. A weighted linear regression showed ORR has significantly increased by 6.0% over time (slope = 0.46% improvement in ORR per year of enrollment, p = 0.003, R² = 0.50; Fig. 3). This finding was preserved in the phase III trials (slope = 0.46% increased ORR/year, p = 0.027, R² = 0.654, 8 studies, 2,084 patients) whereas no linear trend was identified when the analysis was limited to phase II studies (p = 0.72, R² = 0.023, 6 studies, 573 patients).

Thirteen studies (3,021 total patients) contributed data for the median duration of therapy with sorafenib. The median duration of therapy with sorafenib has significantly decreased over time by 53%, from 23.1 weeks to 12.2 weeks since the first clinical trial enrollment in 2005 (p = 0.0037, R² = 0.669) (Fig. 4). When stratified by phase, phase III trials have a significant negative linear trend, with an average reduction of 0.77 weeks of therapy per year of enrollment (p < 0.001, R² = 0.91, 6 studies, 2,397 patients). However, no significant linear trend was noted when this analysis was restricted to phase II trials (p = 0.77, R² = 0.012, 7 studies, 624 patients).

Review of a funnel plot of sample size and median reported OS shows near symmetry around a median OS of 9.9 months with a peak sample size of 578 patients (correlating to the BRISK-FL study) [14]. There is a paucity of smaller studies with prolonged OS (n = 0–300 and median OS 16–20 months). It is unlikely that if these studies have been completed that they would remain unpublished given the potential impact on clinical management of prolonged survivals compared to historical outcomes with sorafenib; however, publication bias cannot be completely ruled out (online suppl. Fig. 1).

To our knowledge, this is the first rigorous review of the clinical outcomes associated with the use of sorafenib in clinical trials of advanced HCC. Several trends are noteworthy. First, the median OS of patients with advanced HCC requiring systemic therapy enrolled on a clinical trial and treated with sorafenib has increased from 2005 to 2019. ORRs by RECIST 1.1 have also significantly increased. During this time, the median duration of treatment with sorafenib has also significantly decreased. Despite this, PFS as measured by RECIST 1.1 has not significantly changed over time. On subgroup analyses, the trends in outcomes appear to be driven mostly by the robust phase III trials, with no discernible trends noted when the analysis was limited to phase II trials. Although the combination of bevacizumab and atezolizumab has since replaced sorafenib as a default frontline therapy for patients with adequate liver function in many parts of the world, these findings have implications for multiple ongoing global phase III studies that continue to use sorafenib as a comparator, and for the clinical management of advanced HCC. For example, the improvement in OS with sorafenib may make the demonstration of superiority over sorafenib more challenging in ongoing clinical studies that are using sorafenib as a control arm. This is exemplified by the failure of nivolumab to demonstrate improved OS versus sorafenib in the CheckMate-459 frontline study, despite producing a median OS of 16.4 months with nivolumab, which compares favorably to the OS of historical sorafenib trials [23]. Currently, there are at least 3 more phase III trials that are ongoing using sorafenib as a control arm – HIMALAYA (NCT03298451) comparing the combination of durvalumab and tremelimumab versus durvalumab versus sorafenib, CheckMate-9DW (NCT04039607) comparing the combination of nivolumab and ipilimumab versus lenvatinib or sorafenib, and COSMIC-312 (NCT03755791) comparing the combination of cabozantanib and atezolizumab versus sorafenib. The trend of increasing median OS of patients receiving sorafenib threatens to affect power calculations and trial accrual and may ultimately affect the outcome and interpretation of these studies. For example, preliminary results were recently announced via press release for COSMIC-312 indicating improved PFS with combination cabozantanib and atezolizumab compared to sorafenib control, but it is unlikely to result in improved OS [26]. It is possible that the failure to demonstrate superior OS in this study was due to better-than-anticipated outcomes with the sorafenib control arm [26]. These preliminary, unpublished results were not included in the present analysis.

It is unclear why median survival with frontline sorafenib has increased over time. The changes do not appear to be driven by quantified patient demographic data such as ECOG PS, age, BCLC stage, prior locoregional therapies, HBV or HCV infection, or CP cirrhosis status, although we do note a positive trend in the proportion of patients with CP A cirrhosis in phase III trials over time. Trial design by phase also does not seem to be a factor, with a significant difference in demographics only detected in a higher proportion of patients with extrahepatic spread in phase III trials compared to phase II trials (online suppl. Table 1). Survival improvements with frontline sorafenib could reflect changes in the use of subsequent therapies, but no second-line therapy was approved for HCC outside of clinical trials until regorafenib’s approval in 2017, whereas the trend of increased survival is evident prior to 2017. Alternatively, these changes in sorafenib treatment outcomes could reflect unquantified changes in patient selection in the context of clinical trials in the 15 years of observations that are reported in this study, such as improvement in multidisciplinary care coordination and supportive care, and innovations in viral hepatitis treatment [9-11]. Such contributing factors to patient outcomes would not be easily captured in the resulting publications [27]. The observation of decreasing duration of therapy with sorafenib in patients on trials in tandem with improving median survival with sorafenib is particularly intriguing and an analysis of post-progression therapies for patients on these studies would be enlightening to determine if patients are discontinuing therapy with sorafenib earlier than anticipated in order to pursue novel second-line treatment options. Lastly, the finding of improvement in ORRs by RECIST 1.1 reinforces the importance of standardized and blinded assessment methods, although the association between blinding and ORR was not formally tested in this analysis.

This analysis has several limitations. The associations we describe here are correlative, determining true causality would require a review of patient level data from each trial. Similarly, it is not known how the observations of this study of improved OS over time correlate with the “real-world” clinical experience over time. Patient selection in oncology trials is a known contributor to the efficacy-effectiveness gap bridging clinical trials to real-world patients, and the outcomes of this efficacy-effectiveness gap have historically been wide in HCC treated with sorafenib [28, 29]. Lastly, we only reviewed published trials of first-line sorafenib in advanced HCC. Although many of the trials are deemed “negative” in that the experimental arm failed to improve the OS compared to the OS with treatment with sorafenib, we cannot rule out the existence of publication bias. However, a review of a funnel plot graphing sample size and median OS revealed an absence of studies of small sample sizes with long median OSs. Given the implication of these hypothetical studies, it is unlikely these studies would remain unpublished.

Although the combination of atezolizumab and bevacizumab is now the standard of care for patients with advanced HCC and preserved liver function, sorafenib remains an option for frontline therapy for patients that do not qualify for atezolizumab and bevacizumab [3, 8]. Here, we demonstrate that the median OS and ORRs of treatment-naive patients with advanced HCC on clinical trials has clearly improved over the last 15 years. Meanwhile, the median duration of therapy has significantly decreased over the same period. The available data do not provide an immediate rationale for these changes and are not accounted for by quantified or published patient demographics. It is possible that better supportive care and multidisciplinary care are contributing to improved survival and the availability of second-line therapies is contributing to decreased durations of therapy with sorafenib. These observations should be considered when designing future clinical trials in advanced HCC.

No human subjects were involved in this study. All included studies required written informed consent and were approved by the relevant Ethics Committees. Please see source studies for more information. This study is exempt from specific Ethical Committee approval as it used only aggregated, publicly available, de-identified data, and posed no risk to human subjects.

M.Y. reports receiving research grants from Incyte, Bristol Myers Squibb, and Exelixis and is a consultant for AstraZeneca, Eisai, Exelixis, and Genentech. The other authors declare no competing interests.

This study was supported by NIH T32CA009679 (T.J.B.), NIH NCI SPORE P50 CA062924 (M.Y.), NIH P30 CA006973 (M.Y.), and NIH UM1CA186691 (M.Y.).

Conception, design, analysis, and interpretation of data were contributed by all authors. Drafting, revision, and final approval were contributed by all authors.

All relevant data are available in this manuscript. Stata source code can be made available on request to corresponding author.

Previous Presentation: Presented in abstract form at the 2021 ASCO Gastrointestinal Cancers Symposium, January 15–17, 2021, Virtual.Twitter Handles: @TimothyJBrownMD @guptaarjun90 @ramsedhom @ShaalanBeg @MarkYarchoan.