Efficacy of First Line Systemic Chemotherapy and Multikinase Inhibitors in Advanced Hepatocellular Carcinoma: A Systematic Review and Network Meta-Analysis

Background Hepatocellular carcinoma (HCC) is the third most fatal cancer, with a 5-year survival rate of 18%. Standard frontline-therapy is multikinase inhibitors (MKIs), but accessibility is still limited, particularly in developing countries. This network meta-analysis (NMA) aimed to compare the efficacy of usual chemotherapy vs MKIs. Method Randomised-controlled trials (RCTs) comparing any among chemotherapy vs MKIs in treatment-naïve patients with advanced HCCs were identified from MEDLINE and SCOPUS databases. Overall survival (OS) and progression-free survival (PFS) probabilities and times were extracted from Kaplan-Meier curves using Digitizer, and then converted to individual patient time-to-event data. A one-stage mixed-effect survival model was applied to estimate median OS and PFS. A two-stage NMA was applied for the overall response rate and adverse events (AEs) outcome. Results A total of 20 RCTs were eligible for NMA. Lenvatinib was the best treatment among single MKIs, with median OS and PFS of 9 and 6.3 months, without significant differences in AEs relative to other MKIs. Median OS and PFS were 0.70 (-0.42, 1.83) and 2.17 (1.41, 2.93) months longer with Lenvatinib than Sorafenib. Among chemotherapy agents, FOLFOX4 had the longest median OS and PFS at 7.9 and 4.3 months, respectively, without significant AEs compared to other chemotherapies. The combination of Sorafenib+Doxorubicin prolonged median OS and PFS to 12.7 and 6.3 months, respectively. Conclusion Use of the MKIs Lenvatinib or Sorafenib as first line systemic treatment for advanced HCC could be beneficial. However, FOLFOX4 might be the optimal choice in a developing country where the health-care budget is limited.


INTRODUCTION
Hepatocellular carcinoma (HCC) is the third most common cause of cancer-related death worldwide (1). The common risk factors are chronic hepatitis B or C virus infection and cirrhosis from any cause. Treatment of HCC depends on the disease stage, which simultaneously considers liver function, performance status, and tumor burden (2,3). Early-stage disease usually requires only local treatment, whereas advanced-stage disease may need Multikinase Inhibitors (MKIs) given preserved liver functions; otherwise, supportive care is the only option.
Sorafenib was the first MKI approved by the Food and Drug Administration (FDA) in 2017 as frontline therapy for advanced-stage HCC. Sorafenib increased median overall survival (OS) to 10.7 months compared with 7.9 months in placebo (4). Other MKIs [i.e., Brivanib (5), Sunitinib (6), and Linifanib (7)] were subsequently tested in phase-III trials, but failed to improve OS relative to Sorafenib, until Lenvatinib, approved by the FDA in 2018, was shown to prolong OS to 13.6 months (8).

METHODS
This SR and NMA were conducted according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and registered to PROSPERO (CRD42019145620). Relevant studies were identified from MEDLINE via PubMed and SCOPUS databases through to 30 th November, 2019. The search terms were constructed based on patients (advanced or unresectable HCC), interventions (i.e., chemotherapy and MKIs), and outcomes of interest (see Table 1). Titles and abstracts were screened by one reviewer (SO) then randomly checked by second reviewer (SR). The full texts were then independently selected by two reviewers (SO and SR), and disagreements were resolved through discussion with a third reviewer (ATh). The most recent studies were selected when there were multiple publications. The reference lists were reviewed to identify additional relevant studies.
RCTs were excluded if they compared systemic with nonmedical treatments (e.g., surgery, liver transplant), local treatments (e.g., chemoembolisation, radiotherapy, and hepatic arterial/portal vein chemo infusion) or non-chemotherapy/MKI (e.g., hormonal treatment, tumor vaccine, or gene therapy). If there were in an untranslatable languages, or had insufficient data for pooling after three contact attempts with authors they were also excluded.
PFS was defined as time since randomization to first occurrence of disease progression or death from cancer-related event. If RCTs (4,5,7,(18)(19)(20)(21) reported only time to progression which was defined as time since randomization to disease progression, time to progression was used instead of PFS. ORR was defined as the proportion of patients who had a best objective tumor response of complete response or partial response. Tumor response was classified using World Health Organisation (WHO) (22) or the response evaluation criteria in solid tumors (RECIST) (23) criteria according to RCTs' report. AEs were at least one of grade 3 or higher of following AEs as for the National Cancer Institute's Common Terminology Criteria for Adverse Events (24): anemia, neutropenia, thrombocytopenia, diarrhea, hand-foot skin reaction, hypertension, or hyponatremia. If RCTs reported these individual AEs rather than overall AEs, the AE with the maximum incidence was used.

Data Extraction and Quality Assessment
Two reviewers (SO and AP) independently extracted the following data: general characteristics (i.e., number of patients, RCT phase, country), patient characteristics (i.e., gender, mean age, percent hepatitis B/C, Child-Pugh A classification, Barcelona Clinic Liver Cancer stage C, portal vein involvement, extrahepatic spreading), treatment regimens, and outcomes (i.e., complete response and partial response, OS, PFS, and AEs).
The probabilities and times of OS and PFS were also extracted from the Kaplan-Meier curve using Digitizer program (25), and then used to simulate individual patient time-to-event data (26). Any disagreements between the two reviewers were resolved by consensus with a third reviewer (ATh). The main outcome of interest was OS; and secondary outcomes were PFS, ORR, and AEs.
Two reviewers (SO and ATa) used the Revised Cochrane Risk of Bias tool (RoB 2) (27) to assess the quality of RCTs; and any disagreements were resolved by consensus with a third reviewer (Ath).

Statistical Analysis
Interventions were coded according to treatment as placebo, single-chemotherapy, combined-chemotherapy, combinedchemotherapy with MKIs, single-MKIs, and combined-MKIs. For individual patient time-to-event data for OS and PFS, a one-stage approach using a mixed-effect parametric survival model (28) was applied to obtain relative treatment effects. Appropriate survival distributions (e.g., Weibull, exponential, log-logistic, log-normal, generalized gamma) were assessed using Akaike's Information Criterion (AIC); and the model with the smallest AIC was used.
For dichotomous outcomes (i.e., ORR and AEs), the relative treatment effect, risk ratio (RR), were compared using a twostage NMA. The probability of being the best treatment in maximizing ORR and minimal AEs was assessed using a rankogram and the Surface Under the Cumulative Ranking curve (SUCRA). Transitivity was explored by comparing the distribution of co-variables (e.g., gender, co-morbidity, Child-Pugh classification, major vascular invasion, extrahepatic spreading) among comparisons. A comparison-adjusted funnel plot was used to assess publication bias.
All analyses were performed with STATA version 16.0 (29). A two-sided p-value less than 0.05 was considered statistically significant for all tests.

Characteristics of Included Studies
A total of 4,662 studies were identified; and reasons for exclusion were reported in the PRISMA flow diagram see Figure 1. The overall risk of bias was low at 75%; see Figure 2. Inter-observer agreement between reviewers was high for both data extraction (kappa = 0.96) and risk of bias assessment (kappa = 0.88).
Twenty RCTs with 7,846 patients were included in the NMA (4-8, 18, 19, 21, 30-40). Most were conducted in Asia (40%), in male patients (76.0% -91.5%), with Child-Pugh A disease except for one study focusing on Child-Pugh B disease (33). The mean age was 49.3 to 65.6 years. A network map for each outcome was constructed see Figure 3, and the characteristics of the included RCTs are shown in more detail in Table 2.

Overall Survival
Nineteen RCTs (4-8, 18, 19, 21, 30, 32-40) assessed OS with 16 treatment regimens, see Figure 3A. Sorafenib+GEMOX was excluded due to the Kaplan-Meier curve not being reported. One-stage NMA with a mixed-effect accelerated failure time model with log-normal distribution see Table 3, and 5000replication bootstrap was applied to estimate the median OS of each treatment, see Table 4. Among single-chemotherapy regimens, the median OS of Doxorubicin was a month longer than Nolatrexed, but this was not significant (6.9 versus 5.9 months), whereas the median OS of PIAF and FOLFOX4 were not much different (i.e., 7.3 versus 7.9 months). Median OS of single-MKI regimens varied from 6.7 to 9.0 months, in which Lenvatinib/Nintedanib and Sunitinib had the longest and shortest median OSs respectively. Combined-MKI regimens showed some additional benefit with Sorafenib+Doxorubicin having a median OS of 12.7 months. All relative treatment effects by median OS were then compared among active treatments and placebo/no-treatment see Table 5. None of the chemotherapy agents was statistically significantly different when compared to placebo, with differences in median OS ranging from -0.2 to 1.8 months. However, six of seven single-MKIs were significantly different to placebo, prolonging median OS by 0.6 to 3.0 months. Among combination regimens, Sorafenib+Erlotinib and Sorafenib+Doxorubicin, showed median OS (95% confidence interval (CI)) longer than placebo by 2.55 (1.04, 4.05) and 6.62 (0.92, 12.33) months, respectively. Comparing  Considering current common treatments (i.e., Doxorubicin, FOLFOX4, Sorafenib, and Lenvatinib), Lenvatinib showed the longest OS followed by Sorafenib, FOLFOX4, and Doxorubicin with median OS times of 9.0, 8.3, 7.9, and 6.9 months, respectively. Lenvatinib and Sorafenib showed significantly longer OSs than Doxorubicin. Among chemotherapy agents, FOLFOX4 had the longest OS, and was statistically significant when compared to Nolatrexed (2.01 months, 95% CI of 0.23, 3.80).   Figure 3B. Median PFS varied from 3.0 to 6.3 months, see Table 4. Comparisons of median PFS indicated all regimens had significantly longer PFSs compared with placebo, except Doxorubicin, Nolatrexed, Dovitinib, Nintedanib, and Sorafenib+Erlotinib, see Table 6.

Risk and Benefit Assessment
Risk (AEs) and benefit (OS) were assessed by estimating the incremental risk-benefit ratio of the treatments recommended by clinical practice guidelines, i.e., Sorafenib, Lenvatinib, and FOLFOX4. Incremental risk-benefit ratios of Lenvatinib and FOLFOX4 versus Sorafenib were jointly simulated for each pair using 1000 Monte Carlo simulations, and then risk of AEs and benefit of OS were plotted on y-and x-axis, respectively, see Figure 4.
For Lenvatinib versus Sorafenib, most AEs and OS fell in the right and the left upper quadrants with symmetry about the xaxis, but not in the y-axis see Figure 4A; indicating that Lenvatinib offered no benefit in terms of OS, but had higher AEs compared to Sorafenib. FOLFOX4 appears to have lower OS and higher AEs than Sorafenib see Figure 4B.

DISCUSSION
We conducted a NMA to assess the efficacy of first-line HCC treatments focusing on chemotherapy agents, and MKIs using data from 20 phase II-III RCTs. Our findings suggested that Lenvatinib and FOLFOX4 were the best of the MKIs and chemotherapy agents respectively, prolonging OS and PFS by 9 and 6.3 months and 7.9 and 4.3 months respectively without significantly different AEs relative to other agents in their class. Combining these classes, i.e. Sorafenib+Doxorubicin could prolong OS and PFS to as long as 12.7 and 6.3 months.
Our NMA included all chemotherapy agents and MKIs which had been studied in RCTs up to 2019, including Lenvatinib (8), which was not included in the previous NMA (9). This allowed us to compare the efficacy between chemotherapy and MKIs by borrowing common comparators (e.g., placebo, Sorafenib), which also had not been done before. The previous NMA (9) indicated that Sorafenib, Linifanib, Brivanib, and Sunitinib were significantly better in prolonging OS than placebo, and that the combination of Sorafenib+Erlotinib was the best in prolonging OS, followed by Sorafenib. Our updated findings indicated that Lenvatinib is better than Sorafenib in prolonging OS, PFS, and showed higher ORR. Considering only treatment regimens recommended by the National Comprehensive Cancer Network (NCCN) (41), including Sorafenib, Lenvatinib, and FOLFOX4, our findings showed that Lenvatinib provides the longest OS and PFS with no significant difference in AEs compared to other treatments.
For chemotherapy agents, only direct meta-analysis of Oxaliplatin-based chemotherapy had been published (12,42). The pooled median PFS for Oxaliplatin-based chemotherapy was similar to our FOLFOX4 regimen (4.2 (12) to 4.7 (42) months vs. 4.3 months) but OS was slightly longer than our results (9.3 (12) to 9.4 (42) months vs 7.9 months). The difference in OS may come from heterogeneity in the combination of chemotherapy with Oxaliplatin.       Although our results favored MKIs as the treatment of choice in advanced HCC, many developing countries may have less access to these drugs due to high cost factors. Therefore, FOLFOX4 may still be considered as the best treatment option in these settings, as suggested by our findings. FOLFOX4 was ranked the highest among chemotherapy agents, but still showed shorter OS and higher AEs rates compared to Sorafenib and Lenvatinib.
The combination of chemotherapy and MKIs (i.e., Sorafenib +Doxorubicin or Sorafenib+GEMOX) or the combination of MKIs (Sorafenib+Erlotinib, Sorafenib+Everolimus, or Bevacizumab+Erlotinib) may also hold some promising outcome for improving OS and PFS. However, all of the combination treatments except Sorafenib+Erlotinib were from phase II RCTs (21,30,31,38), which may overestimate the efficacy due to high selection bias in those trials. Those findings require further evidence for confirmation.
Our study has some strengths. Firstly, we extracted time along with probability of OS/PFS from Kaplan-Meier curves using Digitizer software. Numbers of events (i.e., death and disease progression) along with person-time at each distinct point of the Kaplan-Meier curve were extracted. These data were then converted to individual patient data using methods suggested by Wei and Royston (26). This allowed us to be more flexible in applying a mixed-effect model (28) for time to event data analysis. Median OS and PFS were then estimated for each regimen. Among 20 RCTs, five studies did not report numbers of events and person-time at risk at each distinct time. Therefore, the estimation of hazard functions for these studies might be biased as their hazard ratios, which were estimated from individual level data, were different from their actual reported data.
Nonetheless, there were some limitations, as our review included primary studies published from 1988 to 2019. Since studies of MKIs are more recent than that of chemotherapy, their longer survival times may be confounded with improvements in supportive care over time. Recently, combination immune checkpoint inhibitor (Atezolizumab+Bevacizumab) (43) has been approved as the first-line treatment of advanced HCC and newer treatment options are still being studied in clinical trials. However, accessibility to these treatments is still limited particularly in developing countries due to their costs. Due to the high cost of treatment and grave prognosis of the disease, preventive strategy should be promoted and applied to decrease the incidence of new cases. Adjunctive treatments, which could modify tumorigenesis pathways or major risk factors of HCC such as viral hepatitis B/C infections with epigenetics, microRNAs or microenvironment modification (44,45), should be explored in clinical studies to seek novel therapeutic options for improving survival and quality of life for HCC patients.
In conclusion, for limited resource countries, results from this NMA support the use of MKIs, i.e., Lenvatinib or Sorafenib, as a first-line systemic treatment for advanced HCC with preserved liver function (Child-Pugh A) and non-significant portal vein involvement. FOLFOX4, chemotherapy, might be an option if MKIs are less accessible. This NMA should be updated when more phase-III RCTs are published.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author.