Background

Globally, lung cancer has the highest associated mortality among all malignant cancers [1]. The 5-year survival rate in advanced stage cancers is 15%, as compared to 80% in early stage lung cancers [2]. One of the reasons is that most patients are diagnosed at advanced stages due to lack of sensitive and specific early diagnostic biomarkers [3]. Non-small-cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancers; the remaining 15% cases are classified as small cell lung cancer (SCLC) [4]. Although chemotherapy and targeted therapy are the main clinical treatment especially of stage IV patients, yet there is only 4–5% improvement in 5-year survival rates for stage I-III patients, and no significant improvement for stage IV patients [5]. The diagnostic methods include chest x-ray, computed tomography (CT) and needle biopsy of lung [6, 7]. However, the high cost and/or invasive nature of these investigations limit the widely use in clinical diagnosis.

During past decades, many advances have been made in the identification of tumor-associated markers in body fluids such as plasma, serum or bio-aerosols such as exhaled breath condensate (EBC) [8, 9]which could facilitate early detection and help for treatment monitoring [10]. For lung cancer diagnosis, the leading markers used are carcinoembryonic antigen (CEA), cytokeratin 19 fragments (CYFRA 21–1) and neuron-specific enolase (NSE). CEA, which is closely related to histological classification, is considered valuable for diagnosis of ADC [11]. CYFRA 21–1 and NSE are used for the diagnosis of SCLC [12, 13]. Increasing trend in levels of CEA, CYFRA21-1 and NSE have been associated with metastasis and poor prognosis [14–16]. However, limitations of previous studies are either in small sample sizes (N = 200-300) or not analyzed in combinations.

In this retrospective study we evaluated the predictive values of serum levels of CEA, CYFRA21-1 and NSE for prognosis and occurrence of metastasis, and the association of these biomarkers with clinical characteristics.

Methods

Patients

This study recruited 868 lung cancer patients who were admitted to West China Hospital between 2008 through 2012. All data were obtained from medical records within 2 weeks of diagnosis, and information regarding metastasis was obtained through reports of whole-body CT scan, bone scan, lymph node and fiber optic bronchoscopy biopsy. Survival time was obtained during subsequent follow-up visit or telephonic inquiry. Those patients who did not receive CEA, CYFRA21-1 and NSE determinations and lack of follow-up data were excluded. Data on stage were according to the TNM Classification of Malignant Tumors, 7th Edition [17].

The overall survival time was calculated as time from the date of diagnosis through the date of death or last follow up visit (if the exact date of death was unavailable). Prior to surgery or any other treatments, serum concentrations of CEA, CYFRA21-1 and NSE were measured by immunoassays. Based on the reported literatures, the threshold values for CEA, CYFRA21-1 and NSE levels were 3.4 ng/mL, 3.0 ng/mL and 15.0 ng/mL, respectively [17].

Study design

Depending on the levels of CEA, CYFRA21-1 and NSE, the study subjects were divided into three groups (negative, moderate and high). For CEA analysis, moderate and high groups were defined as 1–10 folds and >10 folds cutoff value, respectively. For CYFRA21-1 analysis 1–3 folds and >3 folds, respectively. For NSE analysis, 1–2 folds and >2 folds, respectively. This analysis was performed in a randomly selected training group (N = 432), reserving the left 436 cases for validation. The cutoff values of three biomarkers for groupings were developed on the training group and tested in a validation group.

Next, we determined the correlations of biomarker levels with three main histological subtypes, ADC, SCC and SCLC. The association analyses of other tumor types (N = 49) such as large cell lung cancers and adenosquamous carcinoma were also performed which showed no positive prognostic value (Data not shown).

Finally, the diagnosis, metastasis and prognostic values of combination patterns of three biomarkers were also evaluated. In brief, patients were grouped as negative, single, double and triple positive of biomarkers. Negative indicated that serum levels of all three biomarkers were below cutoff levels. Single, double, triple positive meant that concentrations of any one, two or all three biomarkers exceeded cutoff values.

Statistical methods

SPSS 19.0 for Windows (SPSS Inc, Chicago, USA) was used for data analyses. Chi-square test was performed to evaluate the inter-group differences. Kaplan-Meier test was used to calculate the survival status of different groups, and Log-rank test was used to compare the survival among three groups. Multivariate Cox regression analysis was used to determine the clinical characteristics, metastasis and survival status in order to estimate the hazards ratio for different serum levels of CEA, CYFRA21-1 and NSE and identify independent predictors of poor prognosis.

Results

Increased levels of CYFRA21-1 significantly correlated with metastatic disease

Total 868 lung cancer patients were randomly divided into training group (TA, 432 cases) and validation (VA, 436 cases) group to confirm the rationality of grouping methods. Among them, 320 patients tested negative (TA: 164, VA: 156) (<3.4 ng/mL) while 365 (TA: 179, VA: 186) and 210 (TA: 89, VA: 94) had moderate and high levels of CEA, respectively. For CYFRA21-1, 231 patients tested negative (TA: 115, VA: 116) while 390 (TA: 190, VA: 200) and 247 (TA: 127, VA: 120) had moderate and high levels, respectively. For NSE, 412 patients (TA: 206, VA: 206) tested negative while 256 (TA: 128, VA 128) and 200 (TA: 98, VA: 102) had moderate and high levels.

The results indicated strong correlations of increased levels of CEA, CYFRA21-1 and NSE with histological classifications in both TA and VA groups (All P < 0.001). CEA and CYFRA21-1 were also related closely to TNM stages in TA and VA groups (P < 0.05, P < 0.01 and P < 0.001), while NSE had dramatic correlation with smoke status (TA: P < 0.01, VA: P < 0.05). CEA correlated closely to bone metastasis (TA: P < 0.05, VA: P < 0.01) and NSE had significant correlation with metastasis of bone (TA: P < 0.001, VA: P < 0.01), liver (TA: P < 0.001, VA: P < 0.01), lymph node (TA: P < 0.01, VA: P < 0.01) and mediastinum (TA: P < 0.01, VA: P < 0.05) (Table 1, Additional file 1: Table S1A and B).

Table 1 The analysis of CYFRA21-1 in all lung cancer patients
Full size table

Among all three biomarkers, levels of CYFRA21-1significantly correlated with occurrence of organ metastasis. Besides metastasis to bone (TA: negative9.6%, moderate 25.3%, high 27.6%, P < 0.01; VA: negative 12.9%, moderate 20.0%, high 34.2%; P < 0.001) and liver (TA: negative 1.7%, moderate10.5%, high 15.6%, P < 0.01; VA: negative 5.2%, moderate11.5%, high 20.0%; P < 0.001), CYFRAY21-1 levels were also associated with metastases to lymph nodes (TA: negative 42.6%, moderate 64.2%, high 70.9%, P < 0.001; VA: negative 50%, moderate 66%, high 65.8%; P < 0.01), pleura (TA: P < 0.01, VA: P < 0.05) and peritoneum (TA: P < 0.01, VA: P < 0.01) (Table 1). However, CEA and NSE levels showed relative poor correlation with metastases (Additional file 1: Table S1A and B), which confirmed the importance of CYFRA21-1 in metastasis. Consistent results between training and validation groups also indicated the grouping rationality although several deviations such as sex, brain metastasis and adrenal gland metastasis in CYFRA21-1 and NSE, while brain and liver metastasis in CEA (Table 1, Additional file 1: Table S1A and B).

Correlation of CYFRA21-1 and NSE with metastases in ADC and SCC, respectively

In this study, the CYFRA21-1 levels showed a stronger correlation with occurrence of metastasis in ADC patients when compared with that of CEA and NSE. It also showed a significant correlation with presence of metastatic lesions in brain (P < 0.05), bone (P < 0.001), liver (P < 0.05), lymph node (P < 0.001), intrapulmonary (P < 0.05), pleural (P < 0.05) and peritoneum (P < 0.05) (Table 2). However, CEA positive levels correlated only with bone (P < 0.05) and liver metastasis (P < 0.05) (Additional file 2: Table S2A), while NSE levels correlated only with metastatic lesions in brain (P < 0.001) and bone (P < 0.001) (Additional file 2: Table S2B).

Table 2 The association analysis between CYFRA21-1 and ADC
Full size table

An interesting finding which differs from those reported earlier is the significant correlation of NSE levels with occurrence of metastasis in SCC patients, as compared with that of CEA and CYFRA21-1. In the present study, NSE levels were associated with occurrence of metastases to brain (P < 0.05), bone (P < 0.05), lymph nodes (P < 0.05), mediastinum (P < 0.05) and peritoneal cavity (P < 0.05) (Table 3). However, CEA levels correlated only with lymph node metastasis (Additional file 3: Table S3A), while CYFRA21-1 was associated with metastasis to brain (Negative: 5.6%; moderate: 2.4%; high: 16.0%, P < 0.05), and lymph node (Negative: 41.7%; moderate: 60%; high: 74.5%; P < 0.05) (Additional file 3: Table S3B).

Table 3 The association analysis between NSE and SCC
Full size table

In the present study, 18.3% of all subjects were small cell lung cancer (SCLC) patients. In these patients, we observed a correlation between increased levels of CEA and occurrence of mediastinal and peritoneal metastasis (P < 0.05) (Additional file 4: Table S4A); between increased levels of CYFRA21-1 and liver metastasis (P < 0.05) (Additional file 4: Table S4B); and between increased NSE levels and occurrence of lymph node metastasis (Negative: 42.1%; moderate: 60.1%; high: 77.8%;P < 0.05) (Additional file 4: Table S4C).

Increased positive numbers of biomarkers as predictors of metastases

The analysis of increased positive numbers of biomarkers in all lung cancer patients was performed in training group and validation groups. In training group, the numbers of negative, single, double and triple groups were 37, 101, 172 and 122 cases, respectively, while 27, 118, 161 and 130 in the validation group. The number TA and VA groups indicated less data deviation among different groups. The results suggested strong correlation of increased positive numbers with stages (TA: P < 0.05, VA: P < 0.05). In metastasis analysis, increased positive numbers related closely to occurrence of metastasis in bone (TA: Neg 10.8%, Single 13.9%, Double 26.2%, Triple 27.9%, P < 0.05; VA: Neg 0%, Single 12.7%, Double 23.6%, Triple 33.1%, P < 0.001) and lymph node (TA: Neg 32.4%, Single 55.4%, Double 59.9%, Triple 73.8%, P < 0.001; VA: Neg 29.6%, Single 50.8%, Double 68.9%, Triple 69.2%, P < 0.001) (Table 4).

Table 4 The analysis of positive numbers of biomarkers in all lung cancer patients
Full size table

The application of 3-tier classification to all types of lung cancers revealed that lymph node metastasis was significantly associated with increased levels of biomarkers (ADC P < 0.05; SCC P < 0.001; SCLC P < 0.05) (Additional file 5: Table S5A-C). In ADC and SCC, increased levels correlated with metastasis to both lymph nodes and other organs (Additional file 5: Table S5A-C).

CYFRA21-1 levels correlated with survival in ADC, SCC and SCLC

Kaplan-Meier survival curves were used to analyze mortality at 3–5 years using SPSS19.0. The results of 3–5 year survival analyses indicated that presence of high concentrations of CEA (TA P < 0.01; VA P < 0.01), CYFRA21-1 (TA P < 0.001; VA P < 0.001), NSE (TA P < 0.05; VA P < 0.05) and positive numbers of biomarkers (TA P < 0.001; VA P < 0.01) were closely associated with survival status in both training group and validation groups (Fig. 1a-d).

Fig. 1
figure 1

The survival status of lung cancer patients in training and validation groups a: CEA, b: CYFRA21-1, c: NSE, d: positive numbers *P < 0.05, **P < 0.001

Full size image

For ADC patients, high levels of CEA (P < 0.001), CYFRA21-1 (P < 0.001), NSE (P < 0.05), and numbers of increased biomarkers (P < 0.001), were all closely associated with survival status of patients (Fig. 2). In SCC patients only CYFRA21-1 was associated with mortality (Additional file 6: Figure S1A). In SCLC patients, the high concentrations of CYFRA21-1 (P < .05) and NSE (P < .05) were closely associated with survival status (Additional file 7: Figure S1B).

Fig. 2
figure 2

The survival functions in ADC patients correlated with different biomarkers *P < 0.05, **P < 0.001

Full size image

Multivariate Cox regression analysis to identify poor prognostic factors

We observed a significant correlation between overall survival and CYFRA21-1, NSE and occurrence of metastasis. Compared with negative group, the hazards ratio increased 1.226 in CYFRA21-1 moderate positive group (Confidence Interval [CI]: 0.977–1.537) and 1.647 in CYFRA21-1 high positive group (CI: 1.273–2.130, P < .001) (Table 5). For NSE, we did not find a significant difference between hazard risk and NSE moderate positive group (HR: 1.010, CI: 0.808–1.263) but the HR increased 1.291 in NSE high positive group compared with that of negative group (CI: 1.032–1.715, P < .05). As expected, occurrence of metastasis was an independent predictor of poor prognosis (HR: 1.291, CI: 1.025–1.625, P < .05) (Table 5).

Table 5 The multivariate analysis of lung cancer patients
Full size table

The specific histological grade analysis indicated that high and moderate levels of serum CYFRA21-1 significantly correlated with poor prognosis (HR: 1.860, CI: 1.036–3.338, P < 0.05) in both ADC and SCLC patients (HR: 1.365, CI: 0.514–3.624, P < 0.05) respectively (Table 6). In SCC and SCLC patients, only occurrence of metastasis was an independent factor for poor prognosis (Table 6). When compared with negative groups, the number of positive biomarkers meant increased mortality risk in SCLC (Single: HR 2.107, CI 0.460–9.644; double: HR 2.247 CI 0.386–13.077; triple: HR 2.508, CI 0.231–27.287) (Table 6) although the associated P value was >0.05.

Table 6 The multivariate analysis of different histological classifications
Full size table

Lung cancer patients were then divided into three groups according to stages (I + II, III and IV) and Cox regression was conducted to analyze which biomarker could act as independent factor of poor prognosis in specific stage. The results indicated that CYFRA21-1 correlated dramatically with poor prognosis in all stages of lung cancer patients (Stages I-II: HR 3.666 CI: 1.095–12.279, P < 0.05; Stage III: HR 1.919 CI: 1.200–3.071, P < 0.05; Stage IV: HR 1.473 CI: 1.056–2.053, P < 0.05) (Table 7 A-C), which confirm the importance of CYFRA21-1 in poor prognosis in different stages of lung cancer besides in specific histological classifications.

Table 7 Cox regression analysis of CEA, CYFRA21-1 and NSE in different stages of lung cancer
Full size table

Discussion

Although several tumor markers for lung cancer have been identified, none of them is specific for lung cancer diagnosis. CYFRA21-1 has been reported as a poor prognostic factor in various cancers, while NSE has been associated with metastasis, and also used for monitoring response to treatment in multiple myeloma. However, these important biomarkers have not been thoroughly investigated in lung cancer patients. In our study, analyses were performed to confirm the correlations between serums CEA, CYFRA 21–1, NSE, as well as the number of positive biomarkers and metastasis and survival status of lung cancer patients.

All patients were randomly divided into training and validation groups to confirm the grouping rationality of this study. In brief, survival curves and associations with clinical characteristics in the validation group were generally similar to those in training group, though there were some non-significant inconsistence in sex and several organs of metastasis. The results indicated that the increased levels of CYFRA21-1 were strongly associated with metastatic sites and histological grades of lung cancers in both training and validation groups. In specific histological subtypes (ADC, SCC and SCLC) analyses, we also found that CYFRA21-1 correlated more closely to metastasis and survival status than CEA and NSE. To our knowledge, these results have not been reported in any of the earlier literatures.

In multivariate Cox regression analysis, only CYFRA21-1 and NSE were found to be independent predictors of prognosis in lung cancer patients. When sub-grouped, only CYFRA21-1was an independent predictor of poor prognosis in ADC (1.86-fold increased risk in high concentration group) and SCLC (1.365-fold increased risk in moderate group) but not CEA and NSE. Finally it was found that CYFRA21-1 could act as independent factor in early (I + II) and advanced stages (III and IV) of lung cancer. Thus, CYFRA21-1 appears to be more important as a prognostic predictor than the other two biomarkers.

Several reports have reported the useful roles of CEA in diagnosis of ADC, CYFRA21-1 in SCC and NSE in SCLC [18–21]. The increased levels of biomarkers are known to correlate with cancer progression, with their sensitivity depending largely on tumor stage and histological classification [22]. In contrast with the previous reports [25], we found no correlation between increased CEA levels and brain metastasis; however, we did observe a correlation with bone, liver, pleural and peritoneal metastases. The inconsistency could be explained by the smaller sample size (approximate N = 300). Research also indicated that high circulating concentrations of CYFRA21-1 and CEA were associated with advanced stages of lung cancer; levels that were two times higher than cutoff value were associated with stage III and IV lung cancer patients [23]. Although CYFRA21-1 appears to be the most sensitive and specific marker for NSCLC [26], its relationship with different histological lung cancers has largely remained unknown. An earlier report suggested that CYFRA was a more sensitive and specific marker for SCC diagnosis and was found to be of prognostic values in patients with recurrent NSCLC receiving gefitinib therapy [27, 28]. In our study, however, high levels of CYFRA21-1 correlated with survival status in ADC and SCLC but not in SCC patients. It also could be used as an independent predictor of poor prognosis in ADC and SCLC patients. Currently, NSE is the most widely used marker for diagnosis and prognosis of SCLC patients [24]. In our study, although positive levels of NSE significantly correlated with survival in SCLC, it did not qualify as an independent predictor for poor prognosis.

Conclusions

We designed this study to evaluate whether positive levels of biomarkers correlate with occurrence of metastasis and poor survival. The retrospective design and cross-sectional nature of our study are limitations that did not permit correlation analysis for all clinic pathological parameters. Our study suggested the important role of CYFRA21-1 in predicting occurrence of metastasis as well as poor prognosis in lung cancer patients. Our results could provide important perspectives for diagnosis, prognosis and management of lung cancer.