Prognostic value of serum CYFRA21-1 and CEA for non-small-cell lung cancer

The aim of the study was to assess the clinical prognostic value of serum cytokeratin 19 fragment (CYFRA21-1) and carcinoembryonic antigen (CEA) for non-small-cell lung cancer (NSCLC) patients. Literatures related to effects of serum CYFRA21-1 and CEA on the prognosis of lung cancer patients were retrieved from databases such as PubMed, Springer Link, Embase, Wanfang, and CNKI. Meta-analysis was carried out using RevMan 5.1 software. Ten literatures involving 1990 NSCLC patients were selected in this study. Total survive estimation merging hazard ratio (HR) in all NSCLC patients with high-level serum CYFRA21-1 was 1.64 (95% CI 1.46–1.84, P < 0.001) and that in all NSCLC patients with high level serum CEA was 1.46 (95% CI 1.28–1.65, P < 0.001). Serum CYFRA21-1 and CEA can be used as prognostic factors of NSCLC patients. Combinative detection of the two indices will be more reliable.


Introduction
Lung cancer is one of the most frequent malignant tumor and its mortality has been in fi rst place among urban population died of malignant tumors [ 1 ]. 85% of lung cancer are non-small cell lung cancer (NSCLC). Although previous studies have demonstrated that postoperative chemotherapy plays a signifi cant role in extending life span of NSCLC patients [ 2,3 ], the 5-year survival is still up to 30-50% [ 4 ]. Tumor makers of lung cancer, including Cytokeratin 19 fragment (CYFRA21-1), carcinoembryonic antigen (CEA), specifi c neurons enolase, CA-125, and squamous cell carcinoma antigen (SCC-Ag) [ 5 ] can be used for therapy detection and prognosis [ 6 ]. CYFRA21-1 is a fragment of CK19 and a primitively expressed epithelial cytokeratin. Although complete cytokeratin molecular usually exists as in dissolvable form, there is solubility CYFRA21-1 in the serum of lung tumor patients. They are acidity (I type) subunit which may be released into serum after tumor metastasis and cytolysis and they are useful serum markers . NSCLC patients (including patients in prophase and advanced stage) with high-level serum CYFRA21-1 are poor prognosis. In other words, high serum CYFRA21-1 level may be a useful noninvasiveness marker to identify NSCLC risk. While it still need careful and abundant clinical studies to be ORIGINAL RESEARCH Prognostic value of serum CYFRA 21-1 and CEA for non-small-cell lung cancer confi rmed [ 7 ]. CEA is a glycoprotein in carcinoembryonic cell surface and is one of the earliest applied tumor markers to detect NSCLC. CEA increases in many tumor tissues such as lung cancer, gastrointestinal neoplasms, breast cancer, carcinoid, and liver cancer and become an independent prognosis factor for NSCLC [ 8 ]. However, singleserum tumor marker is hard to reach ideal level in detecting sensitivity and specifi city of lung cancer.

Literature retrieval
We retrieved literatures focusing on the effect of serum CYFRA21-1 and CEA on NSCLC prognosis in any level through searching databases such as PubMed, Springer Link, Embase, Wanfang, and CNKI (language was limited to English and Chinese). The terms were "CYFRA21-1" or "cytokeratin fragment 21-1" and "CEA" or "carcinoma embryonic antigen" and "NSCLC" or "Non-small cell lung cancer" or "lung adenocarcinoma" or "lung squamous" and "prognosis" and "survival."

Inclusion criteria and exclusion criteria
Inclusion criteria: (1) patients who were diagnosed as NSCLC by pathology; (2) initial distribution of the description was defi ned as operation treatment and other treatments such as chemotherapy, radiotherapy, the best supporting nursing care or the combination of all these methods; (3) serum CYFRA 21-1 levels were detected by ELISA method; (4) studies on the relationship between serum CYFRA21-1, CEA, and NSCLC prognosis in any levels; (5) all data were obtained from original literatures and only restricted at complete indices on NSCLC prognosis such as hazard ratio and 95% CI; (6) all data were analyzed by univariate and multivariate survival analysis; (7) full text can be obtained.
Exclusion criteria: (1) literatures which were repetitive report, with low quality and less information, congress abstract, severe bias, and impossible to be utilized; (2) literatures which only provided abstract but no full text information; (3) literatures which did not detect serum CYFRA21-1 and CEA levels.

Data collection
The following data were collected: (1) gender; (2) age; (3) origin date (the day began to detect serum CYFRA21-1 and CEA); (4) classifi cation based on metastasis tumor Tumor Node metastasis (TNM) standard procedure of the Union for International Cancer Control (UICC) [ 19 ]; (5) histology subgroup [ 18 ]; (6) serum CYFRA21-1 and CEA levels. All NSCLC were histologically classifi ed into fi ve groups, including adenocarcinoma, squamous cell carcinoma, adenosquamous carcinoma, bronchoalveolar carcinoma, and large cell carcinoma. Adenocarcinoma, bronchoalveolar carcinoma, and large cell carcinoma were all classifi ed to other NSCLC since their low amount. Therefore, there were only three histology morphous to be analyzed.

Statistical analysis
We calculated standard error (SE) according to the methods in literature from Parmar et al. [ 20 ] and regarding natural logarithm of Hazard Ratio (HR) and its SE were used to merge the HR. Since reference values of serum CYFRA21-1 and CEA internalized into literatures were different, we regarded those upper reference values as high-level serum CYFRA21-1 and CEA. We carried out homogeneity test for the internalized literatures and utilized fi xed effects models to proceed weighted merger when P > 0.05, but random effects models was used if P < 0.05. The merged HR and its 95% CI were presented as forest plots. Taking SE of internalized HR natural logarithm value as abscissa and the HR natural logarithm value as ordinates, the funnel plot was drawn to describe publication bias. RevMan 5.1 statistics software (The Nordic Cochrane Centre, Cochrane Collaboration, Copenhagen) was utilized to carry out meta-analysis for all the data processing, regarding P < 0.05 as statistical signifi cance.

Literature selection and characteristics
As shown in Figure 1 , 258 literatures from databases such as PubMed, Springer Link, Embase, Wanfang, and CNKI were searched. Among them, 10 literatures [9][10][11][12][13][14][15][16][17][18] were selected according to inclusion criteria and exclusion criteria. The participants were from Czech Republic, Germany, Poland, Republic of Korea, Japan, China, Israel, and France. A total of 1990 patients with age ranging from 19 to 89 and follow-up ranging from 60 to 140 months were from the above countries and the therapies for them were recorded. All the 10 internalized literatures had studied many NSCLC prognosis-related factors such as serum CYFRA21-1, CEA levels, tissue type, pathology staging, differentiation level, age, sex, smoking, and SCC-Ag. Neuron-specifi c enolase (NSE), lactate dehydrogenase, white blood cell count and other biomarkers provided the results of univariate and multivariable analysis. This study selected HR in multivariable analysis to remove the mingle effects of other prognosisrelated factors to serum CYFRA21-1 and CEA level and many NSCLC pathology types. The characteristics of included studies were shown in Table 1 .

Publication bias analysis
The funnel plot showed symmetry, which suggested there were no signifi cant publication bias (Fig. 4 ).

Discussion
In the present study, we included 10 literatures in which the detection of CYFRA21-1 and CEA for NSCLC patients to perform a meta-analysis . The results showed that merged HR of NSCLC patients with high serum CYFRA21-1 level is 1.64, and with high serum CEA level is 1.46, suggesting that patients with high serum CYFRA21-1 level or high CEA level are poor prognosis.
The included 10 studies are from different institutes. Therefore, we think the further analysis for subgroup is questionable. However, each research center classifi ed  tumors according to the standard procedure on metastatic tumor TNM from UICC and the therapies (especially chemotherapy schedule), which provided the best support for the homogenicity of the study . In addition, each included research was carried out by multivariable analysis to eliminate the mingle effects of other prognosis-related factors. In this study, we utilized the median age (64 years) as threshold since age affects the life span of all the study individuals, which may lead to limitation in this study. Furthermore, we selected the studies published in Chinese and English, which may result in selection bias.
Sensitivity and specifi city in NSCLC of single detection of serum CYFRA21-1 or CEA are not enough. Many scholars claimed that combination of detection of these two indices should improve the positive ratio of lung cancer diagnosis and evaluation of prognosis [ 21 ]. Therefore, not a single research could determine the prognosis' signifi cance of tumor markers mentioned in their literatures. Most publication or individual propose that high serum CYFRA21-1 and CEA levels suggesting poor prognosis only based on the effect levels of the measure value to mortality and the reliability for NSCLC prognosis .  Further study will determine how to integrate this information to decide the treatment protocols.
However, there were several limitations that must be taken into account when considering the above-mentioned results: on one hand, tumor depth and nodal status were risk factors for prognosis. However, we did not consider these risk factors in the present study. On the other hand, postoperative adjuvant chemotherapy has demonstrated a clear survival benefi t compared to treatment with surgery alone. However, some included studies failed to provide such information, which might have affected the result.
In conclusion, in the present study, we have determined the signifi cance of CYFRA21-1 and CEA levels to NSCLC poor prognosis. Further, large-sample-size cohort studies focusing on this issue are warranted.

Confl ict of Interest
None declared.