Hepatocellular carcinoma (HCC) accounting for most of primary liver cancer (70%–90%) is the fifth most common cancer diagnosed in male and ninth in female worldwide [1]. According to the GLOBOCAN 2012, an estimated 782,500 new liver cancer cases and 745,500 deaths occurred worldwide during 2012, making it the second most common cause of cancer-related deaths (after lung cancer) [1]. Although advancements have been achieved in the treatments of HCC, the prognosis of HCC patients remains poor with 5-year survival rate ranging from 12% to 23% [2]. Serving as a curative and predominant treatment for HCC, the recurrence rate after resection is approximately 50% at 2 years and 75% at 5 years [3]. Therefore, accurate prediction of survival and early detection of recurrence will be critical for HCC management.

[18F] Fluorodeoxyglucose (FDG) PET is a functional imaging tool by providing metabolic information, which is widely used in the detection of gastrointestinal malignancies, such as gastric cancer, pancreatic cancer, and hepatobiliary cancers [4, 5]. Meanwhile, 18F-FDG PET also plays an important role in gastrointestinal malignancies managements, e.g., staging in pancreatic cancer, evaluation of response to treatments in colorectal cancer, and detection of recurrence and metastases in HCC [6–10]. However, the prognostic value of pretreatment 18F-FDG PET on HCC remains inconclusive, though many studies have investigated its role on HCC patients.

This meta-analysis aims to investigate the prognostic role of pretreatment 18F-FDG PET on HCC patients, in which Tumor SUV/Liver SUV (Tsuv/Lsuv) ratio and Tumor SUV (Tsuv) were used as parameters of 18F-FDG PET, and overall survival (OS) and disease-free survival (DFS) were used as outcomes of HCC patients.

Materials and methods

Search strategy

A systemic search of PubMed, Embase, Cochrane library, and Wanfang databases was performed up to June 2015. The following keywords were used: (“hepatocellular carcinoma” or “liver cancer” or “HCC”) and (“PET” or “18F-fluorodeoxyglucose PET/CT” or “18F-FDG PET”). The search strategy used in PubMed is as follows: “((((hepatocellular carcinoma[Title/Abstract]) OR liver cancer[Title/Abstract]) OR HCC[Title/Abstract])) AND (((18F-fluorodeoxyglucose PET/CT[Title/Abstract]) OR 18F-FDG PET[Title/Abstract]) OR PET[Title/Abstract]).” The references list of retrieved articles was manually screened, in order to gain potential eligible studies.

Selection and exclusion criteria

Studies were included if they fulfill the following criteria: (1) studies investigating the prognostic role of 18F-FDG PET on HCC patients; (2) 18F-FDG PET results were obtained before any treatments; (3) survival results were provided in the original article, such as OS or/and DFS; and (4) relative hazard ratios (HRs) with corresponding 95% confidence intervals (CIs) were available.

Studies were excluded if they met any of the following items: (1) case reports, comment letters, reviews, and duplicates; (2) 18F-FDG PET results were obtained after treatments; (3) the tracer used for PET was not 18F-FDG; (4) other parameters were used rather than common indicators (Tsuv/Lsuv ratio, Tsuv value), such as Tumor SUV/Mediastinum SUV ratio; (5) without survival results, such as OS/DFS; (6) HRs with corresponding CIs were not available; (7) the HRs provided in article were paradoxical obviously; and (8) HRs calculated from available data or Kaplan–Meier were significantly different from the original statistical significance, in terms of P value. If the studies were based on the same origin of population, only the most complete ones were enrolled. Besides, we did not set limitations for language or study sample size during this process.

Data extraction

Two investigators performed data extraction from each potentially included study independently. The extracted data are as follows: the first author, year of publication, origin of population, study sample size, cancer stage, treatments, parameters (Tsuv/Lsuv ratio or Tsuv value), cutoff values for parameters, study endpoints (OS, DFS), HRs with corresponding 95% CIs, HR sources (direct, available data, and Kaplan–Meier curve), and follow-up period. If both univariate and multivariate analyzing results were provided in the same study, then we selected the latter one. When HRs were not provided directly in the article, the total numbers of observed deaths/cancer recurrences and the numbers of samples in each group were extracted to calculate HRs [11]. Besides, we also used Engauge Digitizer version 4.1 (http://sourceforge.net) to read the Kaplan–Meier curves when the data above were not available either; then we calculated the HRs with their corresponding CIs as before [11]. However, we excluded articles which provided paradoxical survival results, and articles in which survival results calculated from Kaplan–Meier curve were significantly different from the original statistical significance, in terms of P value. During this process, discrepancies were resolved by consensus in the meeting organized by a senior investigator.

Statistical analysis

This meta-analysis was performed using Stata 10.0. Pooled HRs with corresponding 95% CIs were used to assess the impact of parameters (Tsuv/Lsuv ratio, Tsuv value) on HCC patients. A combined HR > 1 indicated poor prognosis for patients with high Tsuv/Lsuv ratio or high Tsuv value. The heterogeneity among studies was measured using Cochrane Q test (assessing the P value) and I 2 statistic [12]. If I 2 > 50% or/and P < 0.1, the random-effect model was used; otherwise, the fixed-effect model was used. Both Begg’s and Egger’s tests were used to examine the potential publication biases [13, 14]. All P values were two sided, and P < 0.05 indicates statistical significance.

Methodological assessment

In this meta-analysis, we did not perform methodological assessment, because there is no widely agreed quality for assessing prognostic studies [15]. However, we strictly carried out the inclusion and exclusion criteria during the literature search. Besides, we made sure there was no duplication of data during the data extraction, especially those studies conducted in the same center.

Results

Literature search information

Initially, we identified 766 studies by using the key words provided before in the available databases. Next, we excluded 715 studies by reading the title and abstract, because these studies were not related with our research purpose. Of the remained 51 studies, 29 studies were excluded due to duplicates (n = 10), without enough or correct survival data (n = 7), research purpose unrelated (n = 10), and full-text unavailable (n = 2). Finally, there were 22 studies included in this meta-analysis (nineteen in English, two in Chinese, and one in Korean) [16–37]. Details of the search process are given in Fig. 1.

Fig. 1
figure 1

Flowchart of searching relevant studies used in this meta-analysis

Full size image

Characteristics of included studies

The baseline of included studies is shown in Table 1. Totally, there were twenty-two studies with 1721 patients included in this meta-analysis. The study sample size ranged from 25 to 298 (median number 61). Nineteen of the included studies were conducted in Asia (11 in Korea, 5 in Japan, and 3 in China), and the other three were conducted in Germany, Canada, and Belgium. Regarding treatments, transplantation was used in six studies, transarterial chemoembolization (TACE) was used in four studies, resection was used in five studies, multiple treatment was used in six studies, and no treatment modality was reported in the remained one. There were two kinds of parameters from pretreatment 18F-FDG PET results extracted in this meta-analysis. One parameter was Tsuv/Lsuv ratio, which was used in fourteen studies. The other parameter was Tsuv value, which was used in ten studies. In terms of study endpoints, OS was provided in eighteen studies and DFS was provided in seven studies. The follow-up period was available in twenty-one studies, of which thirteen had the longest follow-up period more than 60 months.

Table 1 Characteristics of included studies in this meta-analysis
Full size table

Meta-analysis for the prognostic value of 18F-FDG PET on OS

In this meta-analysis, OS was analyzed by using both Tsuv/Lsuv ratio and Tsuv value from pretreatment 18F-FDG PET. According to Tsuv/Lsuv ratio, ten studies with 836 patients were included during this analysis. Since heterogeneity was found among these studies (I 2 = 51.4%, P = 0.030), a random-effect model was used to calculate the pooled HR (HR = 2.04; 95% CI 1.50–2.79; P = 0.000) (Fig. 2A). In addition, there were ten studies with 937 patients investigating the prognostic role of Tsuv value on HCC patients. There was heterogeneity among these studies (I 2 = 73.1%, P = 0.000), so a random-effect model was used to calculate the pooled HR (HR = 1.53; 95% CI 1.26–1.87; P = 0.000) (Fig. 2B). These results above suggested that both high Tsuv/Lsuv ratio and high Tsuv value were significantly associated with poor OS, indicating that HCC patients with high Tsuv/Lsuv ratio or high Tsuv value suffered from decreased survival rate.

Fig. 2
figure 2

Forest plots for the prognostic value of pretreatment 18F-FDG PET on overall survival (OS) in HCC patients by parameter of Tsuv/Lsuv ratio (A) and Tsuv value (B)

Full size image

Meta-analysis for the prognostic value of 18F-FDG PET on DFS

Totally, there were seven studies with 532 patients investigating the prognostic impact of pretreatment 18F-FDG PET on DFS, by parameter of Tsuv/Lsuv ratio. There was heterogeneity among these studies, so a random-effect model was used to calculate the pooled HR (HR = 7.17; 95% CI 3.58–14.36; P = 0.000). The results above suggested high Tsuv/Lsuv ratio were significantly associated with poor DFS, indicating that HCC patients with high Tsuv/Lsuv ratio suffered from high tumor recurrence rate. Besides, we also identified only one study investigating the association between Tsuv value and prognosis of HCC patients, in which the HR was 2.03 (95% CI 1.05–3.92; P = 0.036).

Subgroup analysis

Since heterogeneity was found in the meta-analysis with OS and DFS, we therefore conducted subgroup analysis for each of them. According to the median number, the cutoff values for Tsuv/Lsuv ratio and Tsuv value in OS were 1.83 and 4.9, respectively. Despite cutoff values, both high Tsuv/Lsuv ratio and high Tsuv value were associated with poor OS (Fig. 2A, B), and high Tsuv/Lsuv ratio was also associated with poor DFS in HCC patients (Fig. 3). Meanwhile, we found that high Tsuv/Lsuv ratio was associated with poor OS in patients who underwent TACE (HR = 2.08; 95% CI 1.45–2.99; P = 0.000) and transplantation (HR = 4.47; 95% CI 2.23–8.96; P = 0.000), but this association was not found in patients who underwent resection (HR = 1.71; 95% CI 1.00–2.92; P = 0.052) (Table 2). Besides, high Tsuv value was also significantly associated with poor OS in patients who underwent TACE (HR = 5.24; 95% CI 1.54–17.81; P = 0.008) and resection (HR = 2.75; 95% CI 1.10–6.87; P = 0.001) (Table 3). Moreover, high Tsuv/Lsuv ratio was correlated with poor DFS in patients who underwent transplantation (HR = 8.67; 95% CI 3.73–20.19; P = 0.000) and resection (HR = 3.81; 95% CI 1.17–12.41; P = 0.026) (Table 4). However, other subgroup analysis factors, such as study sample size, analysis method, and follow-up period, did not affect the statistical significance from meta-analysis results (Tables 2, 3, 4).

Fig. 3
figure 3

Forest plots for the prognostic value of 18F-FDG PET on disease-free survival (DFS) in HCC patients by parameter of Tsuv/Lsuv ratio

Full size image
Table 2 Subgroup analysis for the prognostic value pretreatment 18F-FDG PET on overall survival (OS) in hepatocellular carcinoma (HCC) patients by parameter of Tsuv/Lsuv ratio
Full size table
Table 3 Subgroup analysis for the prognostic value of pretreatment 18F-FDG PET on overall survival (OS) in hepatocellular carcinoma (HCC) patients by parameter of Tsuv value
Full size table
Table 4 Subgroup analysis for the prognostic value of pretreatment 18F-FDG PET on disease-free survival (DFS) by parameter of Tsuv/Lsuv ratio in hepatocellular carcinoma (HCC) patients
Full size table

Generally, subgroup analysis showed that both high Tsuv/Lsuv ratio and high Tsuv value were significantly associated with poorer prognosis in HCC patients, and this association was not altered by subgroup analysis factors except treatments.

Publication bias

In this meta-analysis, both Begg’s and Egger’s tests were used to examine the potential publication bias. Publication bias was found in the meta-analysis with OS by parameters of Tsuv/Lsuv (P = 0.007, 0.003) and Tsuv (P = 0.004, 0.000), and DFS by parameter of Tsuv (P = 0.072, 0.009).

Discussion

HCC is a lethal malignancy, and its incidence is increasing in the United States [38]. Only 30% of HCC patients are diagnosed at localized stage [38], which are suitable for curative resection or radiofrequency (RFA). However, the survival rate after these treatments remains poor due to high tumor recurrence rate [2, 3]. Till now, the prognostic factors for HCC after treatments are alpha-fetoprotein (AFP) level, immunohistochemical makers (P53, Ki67, CD105, etc.), and pathological features including tumor stage and differentiation [39–41]. However, only part of HCC patients have a significant elevation of AFP level, these biomarkers above are not commonly used in our clinical work, and the pathological features are individual rather than universal. Therefore, searching for practical indicators that can predict survival and detect recurrence after treatments is imperative, which will not only guide the life quality of patients but also allow therapies to be more aggressive.

To the best of our knowledge, this is the first meta-analysis investigating the prognostic value of pretreatment 18F-FDG PET on HCC patients. In the present meta-analysis, we included 22 studies with 1721HCC patients and assessed the prognostic value of parameters from 18F-FDG PET on OS and DFS. Meta-analysis results showed that high Tsuv/Lsuv ratio was significantly associated with both poor survival rate (HR = 2.04; 95% CI 1.50–2.79) and early tumor recurrence rate (HR = 1.53; 95% CI 1.26–1.87), and high Tsuv/Lsuv ratio was also correlated with early tumor recurrence rate (HR = 7.17; 95% CI 3.58–14.36; P = 0.000) in HCC patients. These results suggested that both high Tsuv/Lsuv ratio and high Tsuv value can serve as an indicator of poor survival rate, and high Tsuv/Lsuv ratio can serve as an indicator of early tumor recurrence rate in HCC patients. Meanwhile, subgroup analysis showed that the association above was stable despite study sample size, parameter cutoff value, analytic method, and follow-up period changes, but the association above was unsuitable for patients who underwent resection, in which high Tsuv/Lsuv ratio was not significantly associated with poorer OS (HR = 1.71; 95% CI 1.00–2.92; P = 0.052).

However, this meta-analysis does have some potential limitations. The primary concern is publication biases. Some studies did not provide survival results or provided paradoxical results (not included), some articles were unavailable (published in other databases or could not be downloaded), and some studies achieving negative results were not reported, all these may account for the publication biases in our study. The secondary issue is that the number of included studies for each treatment modality is too small when subgroup analyzed by treatments (e.g., two studies about resection in the analysis with OS and one study about resection in the analysis with DFS by parameter of Tsuv/Lsuv ratio, one study in the analysis with OS by parameter of Tsuv, and some studies did not provide explicit modalities). This issue may make the subgroup analysis results by treatments segmentary. As we all know, publication bias and included studies’ number are vey important factors, which may impact the final meta-analysis results. Therefore, the results from our meta-analysis may be estimation, and more well-designed and prospective studies with large population are required to investigate the prognostic value of pretreatment 18F-FDG PET on HCC patients, especially those with explicit treatment modalities.

In conclusion, we showed that both high Tsuv/Lsuv ratio and high Tsuv value from pretreatment 18F-FDG PET were significantly associated with poor survival rate, and high Tsuv/Lsuv ratio was significantly associated with early tumor recurrence rate in HCC patients. Therefore, pretreatment 18F-FDG PET is a useful tool in predicting the prognosis of HCC patients. As our study has some limitations, more prospectively well-designed and large-scale studies are required to investigate the prognostic value of pretreatment 18F-FDG PET on HCC patients, especially those with explicit treatment modalities.