Surrogate endpoint for overall survival in assessment of adjuvant therapies after curative treatment for hepatocellular carcinoma: a re-analysis of meta-analyses of individual patients’ data

The gold standard endpoint to evaluate the effect of treatment for hepatocellular carcinoma (HCC) is overall survival (OS), but it requires a longer follow-up period to observe. This study aimed to identify whether disease-free survival (DFS) could be used as a surrogate endpoint for OS to assess the efficacy of adjuvant therapies after curative treatment (surgical resection and ablation) for HCC patients. A systematic review was conducted to identify trials about curative treatment combined with or without adjuvant therapies (interferon, IFN; or transarterial chemoembolization, TACE) for HCC. Total of 2211 patients’ data from 17 trials were analyzed. At the individual study level, DFS was strongly correlated to OS (ρ = 0.988 and 0.930, 95% CI: 0.965–0.996 and 0.806–0.976 for the studies comparing Radiofrequency ablation (RFA) + TACE to RFA alone; and for the studies comparing curative treatment + IFN to curative treatment alone, respectively). At the trial level, the effects of treatment on DFS and OS were also strongly correlated to each other (R = 0.815 and 0.854, 95% CI: 0.536–0.934 and 0.621–0.948, respectively). In conclusion, DFS could be used as a potential surrogate endpoint for OS to assess the effect of adjuvant therapies after curative treatment for HCC.


INTRODUCTION
Liver cancer is the second most common cause of cancer-related death, with 782,500 new cases and 745,500 cancer deaths occurred in 2012 worldwide [1]. As the major histological subtype, hepatocellular carcinoma (HCC) occupies 70% to 85% of patients with liver cancers [2]. During the past decades and with advances in diagnosis and treatment, the median survival reaches beyond 5 years [3]. With further exploration of treatment for HCC, overall survival (OS) as a standard endpoint to evaluate the efficacy of treatment is complicated, because it requires a large number of patients and longer follow-up period. It is necessary and meaningful to explore a reliable surrogate endpoint for OS to allow early assessment of treatment for HCC Treatment of HCC has changed greatly within the past decades and become a major research area [3]. For the treatment of HCC, surgical resection, liver transplantation and local ablative therapy are considered to be the curative treatment [4]. Surgical resection is preferred choice for HCC without cirrhosis [3]. HCC patients with normal concentration of bilirubin and no portal hypertension, the probability that survival time reaches 5 years is 70% after surgical resection [5]. Radiofrequency ablation (RFA) is commonly recommended as a first line local ablative therapy for tumors less than 5cm [3]. After RFA, survival in Child-Pugh A patients was 50-75% at 5 years [6,7]. In addition, transarterial chemoembolization (TACE) and interferon (IFN) as the non-curative treatments have been confirmed to improve survival [8,9]. TACE induces objective responses in 35% patients and improves 2-year Meta-Analysis survival [8,10]. It has been confirmed that IFN decreased the rates of tumor recurrence and mortality for hepatitis B virus (HBV) or/and hepatitis C virus (HCV) related HCC [9,11]. Pre-and post-operative antiviral and antiinflammatory treatment with IFN has been shown to reduce early and late tumor recurrence rates in HCC patients with HBV or/and HCV infection [11,12]. In order to find out more effective treatment to improve survival, the combined utilization of adjuvant therapy after curative treatment has attracted increasing attention. Meanwhile, it requires strict evaluation criteria to study the effect of adjuvant therapies.
Overall survival as the gold standard endpoint being used to assess the effect of HCC treatment is reliable. However, it requires a large number of patients and longer follow-up period to estimate significant differences between groups of patients [13]. Utilization of a surrogate endpoint at an early stage in clinical trials could speed up assessment of efficacy and reduce costs. It was defined that OS is the time from randomization to death from any cause, and disease-free survival (DFS) was the time from randomization to the first event (loco-regional, distant recurrence, or death from any cause) after treatment [14]. Recent studies had confirmed that DFS is a valid surrogate endpoint for OS in the clinical trials for the treatment of colon cancer, gastric cancer, and lung cancer [14][15][16]. However, there is still no available surrogate endpoint for OS to assess the efficacy of adjuvant therapies in HCC study. The purpose of our study was to evaluate whether DFS could be used as an early surrogate endpoint in studies involving adjuvant therapies after curative treatment for HCC patients.

RESULTS
This study is based on the individual study data of 2211 patients in 17 studies that were included in 8 meta-analyses. The main characteristics about 7 metaanalyses for RFA + TACE vs. RFA and 1 meta-analysis for curative treatment + IFN vs. curative treatment were summarized (Table 1). After reviewing 116 trials including 10069 patients from 7 meta-analyses, repeated trials and patients, as well as the studies failed to get HRs for OS and DFS were excluded, 7 trials containing 1042 patients were conformed to the inclusion criteria for RFA + TACE vs. RFA ( Table 2). The main characteristics were summarized in Table 2. For curative treatment + IFN vs. curative treatment, 10 trials containing 1169 patients were conformed to the inclusion criteria after elimination of unqualified and duplicate data ( Table 3). The main characteristics were summarized in Table 3. The HRs for OS and DFS were either obtained directly or through the Kaplan-Meier survival curves in these studies.
For treatment with RFA + TACE vs. RFA trials, a total of 7 trials containing 1042 patients were available for analysis. High correlation between the treatment effects on DFS and OS was observed, with a rank-correlation coefficient ρ equaled to 0.988 (95% CI, 0.965-0.996; Figure 1A). The 10 curative treatment-based trials (curative treatment and combination IFN) containing 1169 patients exhibited a high correlation between treatment effects on DFS and OS, with a rank-correlation coefficient ρ equaled to 0.930 (95% CI, 0.806-0.976; Figure 1B).
According to the individual study data from 17 trials, the HRs on the endpoints was appraised. Linear regression analysis was carried out to analyze the correlation between treatment effects on DFS and OS in RFA + TACE vs. RFA trials and curative treatment + IFN vs. curative treatment trials, respectively, and it revealed strong correlation between DFS and OS (Figure 2A, 2B). The correlation coefficient R between the HRs were 0.815 (95% CI 0.536-0.934) for RFA + TACE vs. RFA (Figure 2A), and 0.854 (95% CI 0.621-0.948) for curative treatment + IFN vs. curative treatment ( Figure 2B).
Based on the linear model, as the minimum treatment effect on the surrogate endpoint (DFS), the surrogate threshold effects (STE) is necessary to calculate for predicting a non-zero effect on OS. The STE (based on the estimation error adjusted prediction limits) for RFA + TACE vs. RFA alone corresponded to a DFS HR of 0.33 (for a beneficial treatment) or 0.99 (for a harmful treatment; Figure 2A). In addition, for curative treatment + IFN vs. curative treatment alone, the STE corresponded to a DFS HR of 0.87 (for a beneficial treatment) or 1.44 (for a harmful treatment; Figure 2B).
The prediction results from the leave-one-out crossvalidation analysis showed that for DFS, the observed HRs for OS fell between the limits of the 95% prediction intervals in 6 of the 7 studies for RFA + TACE vs. RFA alone and 9 of the 10 studies for curative treatment + IFN vs. curative treatment alone ( Figure 3A, 3B).

DISCUSSION
Fast progress and improvement in treatment has prolonged survival of patients with HCC. However, the best treatment for patients with different stages of HCC remains to be explored. There is no doubt that the best endpoint to be used in these studies is OS, but it requires a longer follow-up period. It has been reported that the time to progression (TTP), closely related to DFS, could be used as a surrogate endpoint for OS in the clinical trials of advanced HCC [42]. In present study, our results showed that the effects of treatment on DFS and OS were strongly correlated in the HCC patients treated with curative treatment (surgical resection and ablation) combined with or without TACE and IFN. Both the correlation coefficient R (trial-level) and the rank-correlation coefficient ρ (individual-level) are close to 1 between DFS and OS in these patients, thus DFS is a validated surrogate endpoint for OS statistically [43]. These findings suggest that DFS could be used as a surrogate endpoint for OS to reduce the duration and cost in the future clinical studies.
As a reasonable candidate for a surrogate of OS, DFS has been confirmed that it is a surrogate for OS in many tumor types [14][15][16]. Moreover, it has been reported that DFS was the preferred potential surrogate endpoint for small HCC according to a questionnaires survey among clinicians and methodologists [44]. Our study confirmed that DFS could be used as a surrogate endpoint for OS to accelerate assessment of efficacy for adjuvant therapy after curative treatment in HCC. These results were based on a re-analysis of meta-analyses of individual patients' data.
It was reported that surgical resection, liver transplantation, or ablation were curative treatment for HCC patients [3]. Meanwhile, an increasing number of adjuvant therapies were used to treat HCC patients. There was no HCC patient with liver transplantation included in this study because of few published study was consistent with our inclusion criteria. In addition, a lot of studies about adjuvant therapies did not provide the HR, and part of studies was not analyzed by using Kaplan-Meier survival curves and the data couldn't be extracted to calculate the HR. TACE and IFN were commonly used as adjuvant therapies after curative treatment for HCC patients nowadays. The majority of trials about TACE and IFN were conformed to the inclusion criteria. This study analyzed the published data of patients with HCC who received curative RFA combined with TACE vs. RFA alone, and curative  Furthermore, cross-validation results confirmed the accurate prediction of treatment effect on OS is based on the effects on DFS for RFA + TACE vs. RFA and curative treatment + IFN vs. curative treatment. Using the STE obtained from our study to predict a non-zero treatment effect on OS, it would require the upper limit of the CI of the estimated HR to fall below 0.33, or the lower limit to be above 0.99 for RFA + TACE vs. RFA, and the upper limit of the CI of the estimated HR to fall below 0.87, or the lower limit to be above 1.44 for curative treatment + IFN vs. curative treatment. Although, both of them are wide interval for HR, the DFS as a surrogate endpoint for OS remains to be validated and it could be a better choice for the study of adjuvant therapy in HCC.
There are a few limitations to our analysis because it was based on the data extracted from literature, rather than based on the data of each patient directly. Our study included 17 trials from 8 meta-analyses, and the data extracted from Kaplan-Meier survival curves was used to calculate HRs according to a reliable method. Furthermore, with increasing clinical trials of adjuvant therapies for HCC, more work remains to be done to analyze the surrogacy of DFS for OS in upcoming study.
In conclusion, our results suggested that DFS could be used as a surrogate endpoint for OS to allow early assessment of efficacy of adjuvant therapies after curative treatment for HCC patients in future clinical trials.

Search strategy and selection criteria
As presented previously [43], the search strategy was divided into two steps. First, a comprehensive  automated literature search was carried out in Biosis, Embase and PubMed Databases for meta-analyses on randomized controlled trials or retrospective cohort studies that compared RFA plus TACE with RFA alone, and curative treatment plus IFN with curative treatment alone, in HCC patients. The literature search used the terms "curative treatment", "surgical resection", "ablation", "RFA", "TACE", "interferon", and "liver cancer" or "HCC" for studies that were published between January, 2000, and December, 2015. Eligible metaanalyses were included following the flow diagram of the PRISMA (Preferred Reporting items for Systematic Reviews and Meta-analysis) Group (2009

Surrogacy criteria
DFS was defined as the time from randomization to the first event (loco-regional, distant recurrence, or death from any cause) after treatment. OS was defined as the time from randomization to death from any cause [14]. The surrogacy criteria were used by Marc Buyse et al. in this study [45]. The approach was based on the strength of association between the surrogate and the true endpoint (the individual-study-level surrogacy) and between the effects of treatment on the surrogate and the true endpoint (the trial-level surrogacy).

Statistical analysis
Statistical analyses were carried out using the Matlab version R2011a and SPSS version 21.0. For each study, we extracted the data of HR (95% CI) for DFS and OS. If the study did not provide the HRs, we extracted the data from the Kaplan-Meier survival curves and calculated the HRs using the method by Jayne F Tierney et al. [46].
The association between the distribution of the true endpoint (OS) and the surrogate endpoint (DFS) was assessed by a bivariate survival model at the individual-level. To quantify the association between the effect of treatment on OS and the effect of treatment on DFS, a linear regression model was used at the triallevel. Treatment effects were estimated by hazard ratios (HRs). We classified correlation values higher than 0.9 as excellent, higher than 0.75 as very good, higher than 0.5 as good, higher than 0.25 as moderate, and equal to or lower than 0.25 as poor.
On the basis of the linear model at the second stage of the two-stage approach, we calculated the surrogate threshold effect (STE), which was defined as the minimum treatment effect on the surrogate (DFS) necessary to predict a non-zero effect on the true endpoint (OS), i.e. the HR was not equal to 1. A future trial requires an upper limit of a confidence interval for the estimated treatment effect (HR) of the surrogate to fall below the STE to predict a non-zero effect on OS, as described previously [47].
For each meta-analysis, we used a leave-one-out cross-validation approach to assess the prediction accuracy of the surrogate model. Each study was left out once and the linear model was rebuilt with the other studies. This model was then applied to the left-out study and a 95% prediction interval was calculated to compare the predicted and the observed treatment effect on OS.

Authorsʼ contributions
Study concept and design: Xia and Wu; acquisition of data: Wu, Huan, Wang and Wen; analysis and interpretation of data: Wu, Xia, Zhang and Yang; drafting of the manuscript: Wu, Huan and Lau; critical revision of the manuscript for important intellectual content: Xia, Lau and Bie; statistical analysis: Wu and Huan; obtained funding and study supervision: Xia.

CONFLICTS OF INTEREST
Authors who have taken part in this study declared that they do not have anything to disclose regarding funding or conflict of interest with respect to this manuscript.