Prognostic Role of Circulating Tumor Cells in Patients with Pancreatic Cancer : a Meta-analysis

Pancreatic cancer, with a patient survival rate among the worst of any solid cancer, is the fourth leading cause of cancer-related morbidity and mortality in Western countries (Tjensvoll et al., 2013). It leads to an estimated 227, 000 deaths per year worldwide and the 5 year survival rate is only 4.6% due to its early metastasis (Zhou et al., 2011). Most of the pancreatic cancer patients are already at a late stage when diagnosed by imaging and it is hard to make a definite diagnosis with screening methods (Ghaneh et al., 2008). Pancreatic cancer can be screened using Sialic acid Lewis antigen CA 19-9 (CA19-9) and carcinoembryonic antigen (CEA), which are serum-based biomarkers mainly used as auxiliary indicator for early diagnosis, however, their utility are notoriously variable among patients (Kokhanenko et al., 2001). People have made attempt to find a new biomarker to detect the tumor and observe the progression. Circulating tumor cells (CTCs) are tumor cells circulating through normal vessels and capillaries, and neovessles formed by tumor induced angiogenesis (Tjensvoll et al., 2013). When cancer cells metastasis or show invasive property, the first step is to invade vessels


Introduction
Pancreatic cancer, with a patient survival rate among the worst of any solid cancer, is the fourth leading cause of cancer-related morbidity and mortality in Western countries (Tjensvoll et al., 2013). It leads to an estimated 227, 000 deaths per year worldwide and the 5 year survival rate is only 4.6% due to its early metastasis (Zhou et al., 2011). Most of the pancreatic cancer patients are already at a late stage when diagnosed by imaging and it is hard to make a definite diagnosis with screening methods (Ghaneh et al., 2008). Pancreatic cancer can be screened using Sialic acid Lewis antigen CA 19-9 (CA19-9) and carcinoembryonic antigen (CEA), which are serum-based biomarkers mainly used as auxiliary indicator for early diagnosis, however, their utility are notoriously variable among patients (Kokhanenko et al., 2001). People have made attempt to find a new biomarker to detect the tumor and observe the progression.
Circulating tumor cells (CTCs) are tumor cells circulating through normal vessels and capillaries, and neovessles formed by tumor induced angiogenesis (Tjensvoll et al., 2013). When cancer cells metastasis or show invasive property, the first step is to invade vessels Xue-Lei Ma & , Yan-Yan Li & , Jing Zhang & , Jing-Wen Huang, Hong-Yuan Jia, Lei Liu*, Ping Li* and disseminated into peripheral blood, therefore, the detection of pancreatic tumor cells in the peripheral circulation could be used to predict the prognosis of pancreatic cancer, due to their metastatic propensity (Zippelius and Pantel 2000). That is to say, CTCs can provide predictive and prognostic information in terms of disease relapse, and overall survival, as reported in breast cancer; (Cristofanilli et al., 2004;Dong et al., 2012;Tarhan et al., 2013), lung cancer (Hiltermann et al., 2012;Ma et al., 2012;Tarhan et al., 2013), colorectal cancer (Cohen et al., 2008;Uen et al., 2008) and prostate cancer (Goodman et al., 2011;Wang et al., 2011). At present, the common approaches to detect CTCs in patients with pancreatic cancer include: 1) immunological assays using antibodies directed against cell surface antigen; 2) PCR-based molecular assays for tumor-derived DNA or RNA extraction from CTCs; and; 3) technologies based on physical or biological properties of cancer cells (Cen et al., 2012).
With the aim of gaining a better insight into the prognostic value of CTCs in patients with pancreatic cancer, our comprehensive study was conducted by pooling published studies using standard meta-analysis techniques.

Search strategy
PubMed and EMBASE were searched on July 19th, 2013. We retrieved articles with combination of the following key words: circulating tumor cells, CTCs, RT-PCR, CK19 mRNA, CK20 mRNA, CEA mRNA, EpCAM and pancreatic cancer.

Study selection and inclusion/exclusion criteria
Titles and abstracts were reviewed for all searched papers, and full text was perused for potentially eligible studies according to our including criteria. To avoid duplicate data, the most integrated study with the longest follow-up time was included if several published studies with duplicate patients were performed in the same Research Center or if one study contained patients also described in another study. But if different patients were included in two studies of the same Research Center, we included both studies. Similarly, if there are multiple sets of data in one study, such as various sampling time (pre and post), we listed all data as separate ones.
In our meta-analysis, our inclusion criteria are as follows: 1) containing patient cases of pancreatic cancer, 2) measuring the presence of CTCs, 3) studies with data available regarding prognostic role of CTCs in pancreatic cancer patients with survival outcomes such as overall survival (OS) and progression free survival/recurrence free survival (PFS/RFS). Studies were excluded in our study: 1) with duplicate data, 2) lacking key information to calculate log hazard ratio (logHR) and SE (logHR) (SE).

Data extraction
Articles were reviewed independently by two investigators (Xuelei Ma and Jingwen Huang) for data extraction. Any discrepancy was further discussed to reach a consensus. Data was extracted from eligible studies by another two investigators (Yanyan Li and Jing Zhang) independently. The primary data was HR with 95% confidence interval (CI) of survival outcomes, including OS and PFS/RFS. Additional data obtained from the studies included publication year, patients' country, number of patients, histology type, average age of patients, sex ratio (Famale/Male), lymphnode invasion status (negative/positive), tumor differentiation (well/ poor), TNM stage, follow-up time and detection method. The statistical data for calculating logHR and SE were also obtained following Parmar's method, such as HR (95%CI) or the Kaplan-Meier survival curves with p value.

Statistical methods
The logHR and SE were required in our analysis. Some of the original papers provided the logHR and SE directly, while other studies didn't. We utilized available data such as HR with 95%CI, or the p-value for log-rank test with Kaplan-Meier survival curve to calculate them instead. The methods were developed by Parmar (Parmar et al., 1998), Williamson (Williamson et al., 2002) and Tierney (Tierney et al., 2007). Those logHRs and SEs were calculated with the methods above.
As the outcome for analysis was survival in patients, the significant outcome was defined as p<0.05. A combined HR>1 frequently indicated the poorer prognosis in CTCs positive cohort. p<0.10 or I 2 >50% represents existed heterogeneity in combined HRs (Higgins et al., 2003). When homogeneity was fine (p≥0.10, I 2 ≤50%), a fixed effects model was applied to secondary analysis, otherwise, a random effects model was performed. All the above calculations, and publication bias which was evaluated using the Begg's funnel plot, were performed by STATA 11.0 (STATA Corporation, College Station, TX).

Correlation between CTCs and the survival outcome
Overall survival: HRs for OS were available in 2 articles, and the survival curves with P values were extracted in the remaining studies for calculation. The  DOI:http://dx.doi.org/10.7314/APJCP.2014.15.15.6015 Circulating Tumor Cells as a Prognostic Marker in Pancreatic Cancer estimated pooled HR for OS was 1.64 (95%CI 1.39-1.94, p<0.00001), which indicated an increased risk of mortality in CTC positive patients. Only one study give the HR for RFS/DFS directly [24], and HRs and 95%CIs of the rest studies were calculated using the survival curves and the p values. The pooled HR and 95%CI for RFS/DFS was 2.36 (95%CI 1.41-3.96, p<0.00001), which suggested a significant increased risk of disease progression in patients with CTCs positivity (shown in Figure 2).

Europe and Non-Europe countries
Seven studies were from Europe and two were from Japan (Asia). The HRs and 95%CIs for OS and RFS/ DFS in patients in Europe were 1.53 (95%CI 1.25-1.83, p<0.0001) and 1.84 (95%CI 1.31-2.58, p=0.004) respectively (shown in Figure 3 C). While, in the patients in Non-Europe (Japan), the HRs for OS and RFS/DFS were 4.07 and 8.36, respectively (shown in Figure 3 D). The results above and other results were shown in Table  2.

Publication bias
Begg's funnel plot and test was used to examine publication bias. Results of meta-analyses of CTC prediction value for OS and DFS/RFS were p=0.010 and p=0.142, respectively (Figure 4).

Discussion
As we know, it was the first time that a comprehensive and detailed meta-analysis revealed the prognostic role of CTCs for pancreatic cancer. The present meta-analysis is based on a relatively large pool of clinical studies and patients. Here, we identified 9 eligible studies which assessed the prognostic value of CTC detection by CellSearch system as well as the PCR-based molecular assays. This analysis provides coherent evidence that the expression of CTCs detected in the peripheral blood is of fine predictive role in patients with pancreatic cancer. The pooled results are fairly stable and not influenced by the CTC detection method, time point of blood withdrawal, source of patient and other characteristics.
The results of our collective evaluation of the literature on pancreatic cancer suggested that CTCs in the peripheral blood have a predictive value. As a thumb, RR>2 was considered as prognostic biomarker in practical. The combined HR and 95%CI for RFS/PFS meet this standard, while the combined HR and 95%CI for OS suggested a predictive marker but not in practical. These results suggested that CTCs detected in peripheral blood was a promising biomarker to speculate survival outcome, evaluate tumor progression and perform as a drug target.
In addition, the adverse prognostic effect of CTCs detection in the peripheral blood was confirmed throughout all performed subgroup analyses. When we divided the results by different time point of blood samples withdrawal including pre-treatment, intra-treatment and post-treatment, the subgroup analysis suggested that CTCs detected in different sampling time have prognostic value. Compared the pooled HR of different sampling time, the pooled HR for RFS/PFS in post-treatment group was overlaid 2, while the other two time point groups were not. These results showed that post-treatment group showed a better predictive role than other time point groups. The main reason might be that post-treatment time incorporated the pretreatment status of CTCs and intra-treatment tumor cells released by surgery, which meant that CTCs detected at post-treatment time were the most relevant to affect tumor prognosis and patient survival time.
If we divided the results by tumor differentiation, we could easily find that the patients group with more well differentiation patients (well differentiation /poor differentiation>50%) have a higher predictive value than other groups. And if we divided the results by lymph node invasion, we could find that the patients with more lymph node invasion (negative/positive lymph node invasion>25%) also have a higher predictive value. We hypothesized that pancreatic cancer cells with poor differentiation or lymph node invasion are easy to metastasis, so CTCs could widely be detected positive in the poor differentiation and lymph node invasion group. Researchers could not predict the survival time in above state. When we divided the studies by the nation of pooled patients, HR for OS and RFS/PFS have a fine result above 2 in non-European group patients, while the HR OS and RFS/PFS was less than 2 in European patients. It suggests that CTCs have a significant strong predictive value in non-European patients rather than European group.
Though, no heterogeneity has been found in our metaanalysis, differences in the detection methods, sampling time, as well as in demographic or clinic pathologic data of included patients ought to be considered as potential sources of heterogeneity. We divided these studies into several subgroups according to above clinical characteristics. When we grouped the analysis into tumor differentiation (well/poor) >50% and <50% groups, the heterogeneity for OS could shrink as 0% and 0%, which means no heterogeneity. When we grouped the analysis into lymph node invasion (negative/positive) >25% and <25% groups, the heterogeneity for OS could also shrink as 0% and 0%. When we grouped the analysis into Europe and Non-Europe countries groups, the heterogeneity for OS could also shrink as 0% and 0%. This meant that the heterogeneity could be eliminated by the subgroup analysis of tumor differentiation, patient's nation and lymph node status. It indicated that these factors might be the reason of heterogeneity, and we used the subgroup analysis in fix model.
Although, the pool analysis of these studies could support a fine practical role for CTCs analysis in pancreatic cancer, challenges exist in this field because many clinical and technical characteristics could affect the newly detecting method. Firstly, there is growing evidence that the CTCs population is heterogeneous, and co-expression of candidate gene markers and CTCs could be useful for disease monitoring, prediction of survival, and response to therapy. Secondly, some CTCs may have weak or no cytokeratin expression because cell differentiation inducing loss of these antigens can be present during EMT (Hofman et al., 2011). Thirdly, activated leukocytes number is higher in patients with cancer than in controls, and may express markers used to detect CTCs. Last but not the least, we combined the different time point of blood withdraw, as we know, the postoperative sampling time might reflect the most relevant CTCs status, rather than preoperative and intro-treatment CTCs level. (Muller & Schlimok 2000)All these limitation indicated that the detection of CTCs have some false negative and positive result technically. (Kowalewska et al., 2006) Publication bias is a major problem in assessing the validity of clinical research studies. The publication bias for OS group analysis has been found by the Begg's test and funnel plot (Begg and Mazumdar 1994). Begg's test and funnel plot suggested that there was significant publication bias for OS group analysis (p=0.01) and no bias for PFS/RFS group analysis (p=0.142). We attempted to eliminated the publication bias by excluding the article caused the major bias, however, we did not find an article according the criteria. We hypothesized that the publication bias might be inherent in the analysis. One reason aroused the publication bias might be a small number of primary studies, and another reason was different detecting methods, sampling time, tumor differentiation, tumor stages and many other characteristics. In turn, we did not find evidence that publication bias may be significantly influencing our results. It should also be noted that our meta-analysis could not completely exclude biases. For example, positive results tend to be accepted by journals, while negative results are often rejected or even not submitted, which probably introduced bias.
Certain limitation still existed in statics method. Firstly, the multivariate analysis of logHR and SE could only be extracted using direct method, while the logHR and SE extracted from the survival curve and p-value are univariate analysis, these univariate analysis and multivariate analysis were combined together to ensure data integrity. Secondly, another source of biases was related to the eligible data extraction based on the Kaplan-Meier survival curve. LogHR and SE could not be reached reasonably well in a few cases.
In conclusion, the meta-analysis show available evidence supporting the proposition that CTCs detected in peripheral blood have a fine predictive role in pancreatic patients especially on the time point of post-treatment. It may be speculated that CTCs could perform as a fine detecting method in pancreatic cancer.